Effects of Ventral Striatum Lesions on Stimulus-Based versus Action-Based Reinforcement Learning.

PubMed

Rothenhoefer, Kathryn M; Costa, Vincent D; Bartolo, Ramón; Vicario-Feliciano, Raquel; Murray, Elisabeth A; Averbeck, Bruno B

2017-07-19

Learning the values of actions versus stimuli may depend on separable neural circuits. In the current study, we evaluated the performance of rhesus macaques with ventral striatum (VS) lesions on a two-arm bandit task that had randomly interleaved blocks of stimulus-based and action-based reinforcement learning (RL). Compared with controls, monkeys with VS lesions had deficits in learning to select rewarding images but not rewarding actions. We used a RL model to quantify learning and choice consistency and found that, in stimulus-based RL, the VS lesion monkeys were more influenced by negative feedback and had lower choice consistency than controls. Using a Bayesian model to parse the groups' learning strategies, we also found that VS lesion monkeys defaulted to an action-based choice strategy. Therefore, the VS is involved specifically in learning the value of stimuli, not actions. SIGNIFICANCE STATEMENT Reinforcement learning models of the ventral striatum (VS) often assume that it maintains an estimate of state value. This suggests that it plays a general role in learning whether rewards are assigned based on a chosen action or stimulus. In the present experiment, we examined the effects of VS lesions on monkeys' ability to learn that choosing a particular action or stimulus was more likely to lead to reward. We found that VS lesions caused a specific deficit in the monkeys' ability to discriminate between images with different values, whereas their ability to discriminate between actions with different values remained intact. Our results therefore suggest that the VS plays a specific role in learning to select rewarded stimuli. Copyright © 2017 the authors 0270-6474/17/376902-13$15.00/0.

Reinforcement learning in supply chains.

PubMed

Valluri, Annapurna; North, Michael J; Macal, Charles M

2009-10-01

Effective management of supply chains creates value and can strategically position companies. In practice, human beings have been found to be both surprisingly successful and disappointingly inept at managing supply chains. The related fields of cognitive psychology and artificial intelligence have postulated a variety of potential mechanisms to explain this behavior. One of the leading candidates is reinforcement learning. This paper applies agent-based modeling to investigate the comparative behavioral consequences of three simple reinforcement learning algorithms in a multi-stage supply chain. For the first time, our findings show that the specific algorithm that is employed can have dramatic effects on the results obtained. Reinforcement learning is found to be valuable in multi-stage supply chains with several learning agents, as independent agents can learn to coordinate their behavior. However, learning in multi-stage supply chains using these postulated approaches from cognitive psychology and artificial intelligence take extremely long time periods to achieve stability which raises questions about their ability to explain behavior in real supply chains. The fact that it takes thousands of periods for agents to learn in this simple multi-agent setting provides new evidence that real world decision makers are unlikely to be using strict reinforcement learning in practice.

Robust reinforcement learning.

PubMed

Morimoto, Jun; Doya, Kenji

2005-02-01

This letter proposes a new reinforcement learning (RL) paradigm that explicitly takes into account input disturbance as well as modeling errors. The use of environmental models in RL is quite popular for both offline learning using simulations and for online action planning. However, the difference between the model and the real environment can lead to unpredictable, and often unwanted, results. Based on the theory of H(infinity) control, we consider a differential game in which a "disturbing" agent tries to make the worst possible disturbance while a "control" agent tries to make the best control input. The problem is formulated as finding a min-max solution of a value function that takes into account the amount of the reward and the norm of the disturbance. We derive online learning algorithms for estimating the value function and for calculating the worst disturbance and the best control in reference to the value function. We tested the paradigm, which we call robust reinforcement learning (RRL), on the control task of an inverted pendulum. In the linear domain, the policy and the value function learned by online algorithms coincided with those derived analytically by the linear H(infinity) control theory. For a fully nonlinear swing-up task, RRL achieved robust performance with changes in the pendulum weight and friction, while a standard reinforcement learning algorithm could not deal with these changes. We also applied RRL to the cart-pole swing-up task, and a robust swing-up policy was acquired.

Reinforcement learning in computer vision

NASA Astrophysics Data System (ADS)

Bernstein, A. V.; Burnaev, E. V.

2018-04-01

Nowadays, machine learning has become one of the basic technologies used in solving various computer vision tasks such as feature detection, image segmentation, object recognition and tracking. In many applications, various complex systems such as robots are equipped with visual sensors from which they learn state of surrounding environment by solving corresponding computer vision tasks. Solutions of these tasks are used for making decisions about possible future actions. It is not surprising that when solving computer vision tasks we should take into account special aspects of their subsequent application in model-based predictive control. Reinforcement learning is one of modern machine learning technologies in which learning is carried out through interaction with the environment. In recent years, Reinforcement learning has been used both for solving such applied tasks as processing and analysis of visual information, and for solving specific computer vision problems such as filtering, extracting image features, localizing objects in scenes, and many others. The paper describes shortly the Reinforcement learning technology and its use for solving computer vision problems.

Reinforcement Learning Trees

PubMed Central

Zhu, Ruoqing; Zeng, Donglin; Kosorok, Michael R.

2015-01-01

In this paper, we introduce a new type of tree-based method, reinforcement learning trees (RLT), which exhibits significantly improved performance over traditional methods such as random forests (Breiman, 2001) under high-dimensional settings. The innovations are three-fold. First, the new method implements reinforcement learning at each selection of a splitting variable during the tree construction processes. By splitting on the variable that brings the greatest future improvement in later splits, rather than choosing the one with largest marginal effect from the immediate split, the constructed tree utilizes the available samples in a more efficient way. Moreover, such an approach enables linear combination cuts at little extra computational cost. Second, we propose a variable muting procedure that progressively eliminates noise variables during the construction of each individual tree. The muting procedure also takes advantage of reinforcement learning and prevents noise variables from being considered in the search for splitting rules, so that towards terminal nodes, where the sample size is small, the splitting rules are still constructed from only strong variables. Last, we investigate asymptotic properties of the proposed method under basic assumptions and discuss rationale in general settings. PMID:26903687

Effective reinforcement learning following cerebellar damage requires a balance between exploration and motor noise

PubMed Central

Therrien, Amanda S.; Wolpert, Daniel M.

2016-01-01

Abstract See Miall and Galea (doi: 10.1093/awv343 ) for a scientific commentary on this article. Reinforcement and error-based processes are essential for motor learning, with the cerebellum thought to be required only for the error-based mechanism. Here we examined learning and retention of a reaching skill under both processes. Control subjects learned similarly from reinforcement and error-based feedback, but showed much better retention under reinforcement. To apply reinforcement to cerebellar patients, we developed a closed-loop reinforcement schedule in which task difficulty was controlled based on recent performance. This schedule produced substantial learning in cerebellar patients and controls. Cerebellar patients varied in their learning under reinforcement but fully retained what was learned. In contrast, they showed complete lack of retention in error-based learning. We developed a mechanistic model of the reinforcement task and found that learning depended on a balance between exploration variability and motor noise. While the cerebellar and control groups had similar exploration variability, the patients had greater motor noise and hence learned less. Our results suggest that cerebellar damage indirectly impairs reinforcement learning by increasing motor noise, but does not interfere with the reinforcement mechanism itself. Therefore, reinforcement can be used to learn and retain novel skills, but optimal reinforcement learning requires a balance between exploration variability and motor noise. PMID:26626368

Effective reinforcement learning following cerebellar damage requires a balance between exploration and motor noise.

PubMed

Therrien, Amanda S; Wolpert, Daniel M; Bastian, Amy J

2016-01-01

Reinforcement and error-based processes are essential for motor learning, with the cerebellum thought to be required only for the error-based mechanism. Here we examined learning and retention of a reaching skill under both processes. Control subjects learned similarly from reinforcement and error-based feedback, but showed much better retention under reinforcement. To apply reinforcement to cerebellar patients, we developed a closed-loop reinforcement schedule in which task difficulty was controlled based on recent performance. This schedule produced substantial learning in cerebellar patients and controls. Cerebellar patients varied in their learning under reinforcement but fully retained what was learned. In contrast, they showed complete lack of retention in error-based learning. We developed a mechanistic model of the reinforcement task and found that learning depended on a balance between exploration variability and motor noise. While the cerebellar and control groups had similar exploration variability, the patients had greater motor noise and hence learned less. Our results suggest that cerebellar damage indirectly impairs reinforcement learning by increasing motor noise, but does not interfere with the reinforcement mechanism itself. Therefore, reinforcement can be used to learn and retain novel skills, but optimal reinforcement learning requires a balance between exploration variability and motor noise. © The Author (2015). Published by Oxford University Press on behalf of the Guarantors of Brain.

Can model-free reinforcement learning explain deontological moral judgments?

PubMed

Ayars, Alisabeth

2016-05-01

Dual-systems frameworks propose that moral judgments are derived from both an immediate emotional response, and controlled/rational cognition. Recently Cushman (2013) proposed a new dual-system theory based on model-free and model-based reinforcement learning. Model-free learning attaches values to actions based on their history of reward and punishment, and explains some deontological, non-utilitarian judgments. Model-based learning involves the construction of a causal model of the world and allows for far-sighted planning; this form of learning fits well with utilitarian considerations that seek to maximize certain kinds of outcomes. I present three concerns regarding the use of model-free reinforcement learning to explain deontological moral judgment. First, many actions that humans find aversive from model-free learning are not judged to be morally wrong. Moral judgment must require something in addition to model-free learning. Second, there is a dearth of evidence for central predictions of the reinforcement account-e.g., that people with different reinforcement histories will, all else equal, make different moral judgments. Finally, to account for the effect of intention within the framework requires certain assumptions which lack support. These challenges are reasonable foci for future empirical/theoretical work on the model-free/model-based framework. Copyright © 2016 Elsevier B.V. All rights reserved.

Fuzzy Q-Learning for Generalization of Reinforcement Learning

NASA Technical Reports Server (NTRS)

Berenji, Hamid R.

1996-01-01

Fuzzy Q-Learning, introduced earlier by the author, is an extension of Q-Learning into fuzzy environments. GARIC is a methodology for fuzzy reinforcement learning. In this paper, we introduce GARIC-Q, a new method for doing incremental Dynamic Programming using a society of intelligent agents which are controlled at the top level by Fuzzy Q-Learning and at the local level, each agent learns and operates based on GARIC. GARIC-Q improves the speed and applicability of Fuzzy Q-Learning through generalization of input space by using fuzzy rules and bridges the gap between Q-Learning and rule based intelligent systems.

Generalization of value in reinforcement learning by humans.

PubMed

Wimmer, G Elliott; Daw, Nathaniel D; Shohamy, Daphna

2012-04-01

Research in decision-making has focused on the role of dopamine and its striatal targets in guiding choices via learned stimulus-reward or stimulus-response associations, behavior that is well described by reinforcement learning theories. However, basic reinforcement learning is relatively limited in scope and does not explain how learning about stimulus regularities or relations may guide decision-making. A candidate mechanism for this type of learning comes from the domain of memory, which has highlighted a role for the hippocampus in learning of stimulus-stimulus relations, typically dissociated from the role of the striatum in stimulus-response learning. Here, we used functional magnetic resonance imaging and computational model-based analyses to examine the joint contributions of these mechanisms to reinforcement learning. Humans performed a reinforcement learning task with added relational structure, modeled after tasks used to isolate hippocampal contributions to memory. On each trial participants chose one of four options, but the reward probabilities for pairs of options were correlated across trials. This (uninstructed) relationship between pairs of options potentially enabled an observer to learn about option values based on experience with the other options and to generalize across them. We observed blood oxygen level-dependent (BOLD) activity related to learning in the striatum and also in the hippocampus. By comparing a basic reinforcement learning model to one augmented to allow feedback to generalize between correlated options, we tested whether choice behavior and BOLD activity were influenced by the opportunity to generalize across correlated options. Although such generalization goes beyond standard computational accounts of reinforcement learning and striatal BOLD, both choices and striatal BOLD activity were better explained by the augmented model. Consistent with the hypothesized role for the hippocampus in this generalization, functional

A Reinforcement-Based Learning Paradigm Increases Anatomical Learning and Retention-A Neuroeducation Study.

PubMed

Anderson, Sarah J; Hecker, Kent G; Krigolson, Olave E; Jamniczky, Heather A

2018-01-01

In anatomy education, a key hurdle to engaging in higher-level discussion in the classroom is recognizing and understanding the extensive terminology used to identify and describe anatomical structures. Given the time-limited classroom environment, seeking methods to impart this foundational knowledge to students in an efficient manner is essential. Just-in-Time Teaching (JiTT) methods incorporate pre-class exercises (typically online) meant to establish foundational knowledge in novice learners so subsequent instructor-led sessions can focus on deeper, more complex concepts. Determining how best do we design and assess pre-class exercises requires a detailed examination of learning and retention in an applied educational context. Here we used electroencephalography (EEG) as a quantitative dependent variable to track learning and examine the efficacy of JiTT activities to teach anatomy. Specifically, we examined changes in the amplitude of the N250 and reward positivity event-related brain potential (ERP) components alongside behavioral performance as novice students participated in a series of computerized reinforcement-based learning modules to teach neuroanatomical structures. We found that as students learned to identify anatomical structures, the amplitude of the N250 increased and reward positivity amplitude decreased in response to positive feedback. Both on a retention and transfer exercise when learners successfully remembered and translated their knowledge to novel images, the amplitude of the reward positivity remained decreased compared to early learning. Our findings suggest ERPs can be used as a tool to track learning, retention, and transfer of knowledge and that employing the reinforcement learning paradigm is an effective educational approach for developing anatomical expertise.

Behavioral and neural properties of social reinforcement learning

PubMed Central

Jones, Rebecca M.; Somerville, Leah H.; Li, Jian; Ruberry, Erika J.; Libby, Victoria; Glover, Gary; Voss, Henning U.; Ballon, Douglas J.; Casey, BJ

2011-01-01

Social learning is critical for engaging in complex interactions with other individuals. Learning from positive social exchanges, such as acceptance from peers, may be similar to basic reinforcement learning. We formally test this hypothesis by developing a novel paradigm that is based upon work in non-human primates and human imaging studies of reinforcement learning. The probability of receiving positive social reinforcement from three distinct peers was parametrically manipulated while brain activity was recorded in healthy adults using event-related functional magnetic resonance imaging (fMRI). Over the course of the experiment, participants responded more quickly to faces of peers who provided more frequent positive social reinforcement, and rated them as more likeable. Modeling trial-by-trial learning showed ventral striatum and orbital frontal cortex activity correlated positively with forming expectations about receiving social reinforcement. Rostral anterior cingulate cortex activity tracked positively with modulations of expected value of the cues (peers). Together, the findings across three levels of analysis - social preferences, response latencies and modeling neural responses – are consistent with reinforcement learning theory and non-human primate electrophysiological studies of reward. This work highlights the fundamental influence of acceptance by one’s peers in altering subsequent behavior. PMID:21917787

Somatic and Reinforcement-Based Plasticity in the Initial Stages of Human Motor Learning.

PubMed

Sidarta, Ananda; Vahdat, Shahabeddin; Bernardi, Nicolò F; Ostry, David J

2016-11-16

As one learns to dance or play tennis, the desired somatosensory state is typically unknown. Trial and error is important as motor behavior is shaped by successful and unsuccessful movements. As an experimental model, we designed a task in which human participants make reaching movements to a hidden target and receive positive reinforcement when successful. We identified somatic and reinforcement-based sources of plasticity on the basis of changes in functional connectivity using resting-state fMRI before and after learning. The neuroimaging data revealed reinforcement-related changes in both motor and somatosensory brain areas in which a strengthening of connectivity was related to the amount of positive reinforcement during learning. Areas of prefrontal cortex were similarly altered in relation to reinforcement, with connectivity between sensorimotor areas of putamen and the reward-related ventromedial prefrontal cortex strengthened in relation to the amount of successful feedback received. In other analyses, we assessed connectivity related to changes in movement direction between trials, a type of variability that presumably reflects exploratory strategies during learning. We found that connectivity in a network linking motor and somatosensory cortices increased with trial-to-trial changes in direction. Connectivity varied as well with the change in movement direction following incorrect movements. Here the changes were observed in a somatic memory and decision making network involving ventrolateral prefrontal cortex and second somatosensory cortex. Our results point to the idea that the initial stages of motor learning are not wholly motor but rather involve plasticity in somatic and prefrontal networks related both to reward and exploration. In the initial stages of motor learning, the placement of the limbs is learned primarily through trial and error. In an experimental analog, participants make reaching movements to a hidden target and receive positive

Somatic and Reinforcement-Based Plasticity in the Initial Stages of Human Motor Learning

PubMed Central

Sidarta, Ananda; Vahdat, Shahabeddin; Bernardi, Nicolò F.

2016-01-01

As one learns to dance or play tennis, the desired somatosensory state is typically unknown. Trial and error is important as motor behavior is shaped by successful and unsuccessful movements. As an experimental model, we designed a task in which human participants make reaching movements to a hidden target and receive positive reinforcement when successful. We identified somatic and reinforcement-based sources of plasticity on the basis of changes in functional connectivity using resting-state fMRI before and after learning. The neuroimaging data revealed reinforcement-related changes in both motor and somatosensory brain areas in which a strengthening of connectivity was related to the amount of positive reinforcement during learning. Areas of prefrontal cortex were similarly altered in relation to reinforcement, with connectivity between sensorimotor areas of putamen and the reward-related ventromedial prefrontal cortex strengthened in relation to the amount of successful feedback received. In other analyses, we assessed connectivity related to changes in movement direction between trials, a type of variability that presumably reflects exploratory strategies during learning. We found that connectivity in a network linking motor and somatosensory cortices increased with trial-to-trial changes in direction. Connectivity varied as well with the change in movement direction following incorrect movements. Here the changes were observed in a somatic memory and decision making network involving ventrolateral prefrontal cortex and second somatosensory cortex. Our results point to the idea that the initial stages of motor learning are not wholly motor but rather involve plasticity in somatic and prefrontal networks related both to reward and exploration. SIGNIFICANCE STATEMENT In the initial stages of motor learning, the placement of the limbs is learned primarily through trial and error. In an experimental analog, participants make reaching movements to a hidden target

Self-Paced Prioritized Curriculum Learning With Coverage Penalty in Deep Reinforcement Learning.

PubMed

Ren, Zhipeng; Dong, Daoyi; Li, Huaxiong; Chen, Chunlin; Zhipeng Ren; Daoyi Dong; Huaxiong Li; Chunlin Chen; Dong, Daoyi; Li, Huaxiong; Chen, Chunlin; Ren, Zhipeng

2018-06-01

In this paper, a new training paradigm is proposed for deep reinforcement learning using self-paced prioritized curriculum learning with coverage penalty. The proposed deep curriculum reinforcement learning (DCRL) takes the most advantage of experience replay by adaptively selecting appropriate transitions from replay memory based on the complexity of each transition. The criteria of complexity in DCRL consist of self-paced priority as well as coverage penalty. The self-paced priority reflects the relationship between the temporal-difference error and the difficulty of the current curriculum for sample efficiency. The coverage penalty is taken into account for sample diversity. With comparison to deep Q network (DQN) and prioritized experience replay (PER) methods, the DCRL algorithm is evaluated on Atari 2600 games, and the experimental results show that DCRL outperforms DQN and PER on most of these games. More results further show that the proposed curriculum training paradigm of DCRL is also applicable and effective for other memory-based deep reinforcement learning approaches, such as double DQN and dueling network. All the experimental results demonstrate that DCRL can achieve improved training efficiency and robustness for deep reinforcement learning.

Autonomous reinforcement learning with experience replay.

PubMed

Wawrzyński, Paweł; Tanwani, Ajay Kumar

2013-05-01

This paper considers the issues of efficiency and autonomy that are required to make reinforcement learning suitable for real-life control tasks. A real-time reinforcement learning algorithm is presented that repeatedly adjusts the control policy with the use of previously collected samples, and autonomously estimates the appropriate step-sizes for the learning updates. The algorithm is based on the actor-critic with experience replay whose step-sizes are determined on-line by an enhanced fixed point algorithm for on-line neural network training. An experimental study with simulated octopus arm and half-cheetah demonstrates the feasibility of the proposed algorithm to solve difficult learning control problems in an autonomous way within reasonably short time. Copyright © 2012 Elsevier Ltd. All rights reserved.

Knowledge-Based Reinforcement Learning for Data Mining

NASA Astrophysics Data System (ADS)

Kudenko, Daniel; Grzes, Marek

Data Mining is the process of extracting patterns from data. Two general avenues of research in the intersecting areas of agents and data mining can be distinguished. The first approach is concerned with mining an agent’s observation data in order to extract patterns, categorize environment states, and/or make predictions of future states. In this setting, data is normally available as a batch, and the agent’s actions and goals are often independent of the data mining task. The data collection is mainly considered as a side effect of the agent’s activities. Machine learning techniques applied in such situations fall into the class of supervised learning. In contrast, the second scenario occurs where an agent is actively performing the data mining, and is responsible for the data collection itself. For example, a mobile network agent is acquiring and processing data (where the acquisition may incur a certain cost), or a mobile sensor agent is moving in a (perhaps hostile) environment, collecting and processing sensor readings. In these settings, the tasks of the agent and the data mining are highly intertwined and interdependent (or even identical). Supervised learning is not a suitable technique for these cases. Reinforcement Learning (RL) enables an agent to learn from experience (in form of reward and punishment for explorative actions) and adapt to new situations, without a teacher. RL is an ideal learning technique for these data mining scenarios, because it fits the agent paradigm of continuous sensing and acting, and the RL agent is able to learn to make decisions on the sampling of the environment which provides the data. Nevertheless, RL still suffers from scalability problems, which have prevented its successful use in many complex real-world domains. The more complex the tasks, the longer it takes a reinforcement learning algorithm to converge to a good solution. For many real-world tasks, human expert knowledge is available. For example, human

A Reinforcement-Based Learning Paradigm Increases Anatomical Learning and Retention—A Neuroeducation Study

PubMed Central

Anderson, Sarah J.; Hecker, Kent G.; Krigolson, Olave E.; Jamniczky, Heather A.

2018-01-01

In anatomy education, a key hurdle to engaging in higher-level discussion in the classroom is recognizing and understanding the extensive terminology used to identify and describe anatomical structures. Given the time-limited classroom environment, seeking methods to impart this foundational knowledge to students in an efficient manner is essential. Just-in-Time Teaching (JiTT) methods incorporate pre-class exercises (typically online) meant to establish foundational knowledge in novice learners so subsequent instructor-led sessions can focus on deeper, more complex concepts. Determining how best do we design and assess pre-class exercises requires a detailed examination of learning and retention in an applied educational context. Here we used electroencephalography (EEG) as a quantitative dependent variable to track learning and examine the efficacy of JiTT activities to teach anatomy. Specifically, we examined changes in the amplitude of the N250 and reward positivity event-related brain potential (ERP) components alongside behavioral performance as novice students participated in a series of computerized reinforcement-based learning modules to teach neuroanatomical structures. We found that as students learned to identify anatomical structures, the amplitude of the N250 increased and reward positivity amplitude decreased in response to positive feedback. Both on a retention and transfer exercise when learners successfully remembered and translated their knowledge to novel images, the amplitude of the reward positivity remained decreased compared to early learning. Our findings suggest ERPs can be used as a tool to track learning, retention, and transfer of knowledge and that employing the reinforcement learning paradigm is an effective educational approach for developing anatomical expertise. PMID:29467638

On the integration of reinforcement learning and approximate reasoning for control

NASA Technical Reports Server (NTRS)

Berenji, Hamid R.

1991-01-01

The author discusses the importance of strengthening the knowledge representation characteristic of reinforcement learning techniques using methods such as approximate reasoning. The ARIC (approximate reasoning-based intelligent control) architecture is an example of such a hybrid approach in which the fuzzy control rules are modified (fine-tuned) using reinforcement learning. ARIC also demonstrates that it is possible to start with an approximately correct control knowledge base and learn to refine this knowledge through further experience. On the other hand, techniques such as the TD (temporal difference) algorithm and Q-learning establish stronger theoretical foundations for their use in adaptive control and also in stability analysis of hybrid reinforcement learning and approximate reasoning-based controllers.

Reinforcement learning in scheduling

NASA Technical Reports Server (NTRS)

Dietterich, Tom G.; Ok, Dokyeong; Zhang, Wei; Tadepalli, Prasad

1994-01-01

The goal of this research is to apply reinforcement learning methods to real-world problems like scheduling. In this preliminary paper, we show that learning to solve scheduling problems such as the Space Shuttle Payload Processing and the Automatic Guided Vehicle (AGV) scheduling can be usefully studied in the reinforcement learning framework. We discuss some of the special challenges posed by the scheduling domain to these methods and propose some possible solutions we plan to implement.

Pragmatically Framed Cross-Situational Noun Learning Using Computational Reinforcement Models

PubMed Central

Najnin, Shamima; Banerjee, Bonny

2018-01-01

Cross-situational learning and social pragmatic theories are prominent mechanisms for learning word meanings (i.e., word-object pairs). In this paper, the role of reinforcement is investigated for early word-learning by an artificial agent. When exposed to a group of speakers, the agent comes to understand an initial set of vocabulary items belonging to the language used by the group. Both cross-situational learning and social pragmatic theory are taken into account. As social cues, joint attention and prosodic cues in caregiver's speech are considered. During agent-caregiver interaction, the agent selects a word from the caregiver's utterance and learns the relations between that word and the objects in its visual environment. The “novel words to novel objects” language-specific constraint is assumed for computing rewards. The models are learned by maximizing the expected reward using reinforcement learning algorithms [i.e., table-based algorithms: Q-learning, SARSA, SARSA-λ, and neural network-based algorithms: Q-learning for neural network (Q-NN), neural-fitted Q-network (NFQ), and deep Q-network (DQN)]. Neural network-based reinforcement learning models are chosen over table-based models for better generalization and quicker convergence. Simulations are carried out using mother-infant interaction CHILDES dataset for learning word-object pairings. Reinforcement is modeled in two cross-situational learning cases: (1) with joint attention (Attentional models), and (2) with joint attention and prosodic cues (Attentional-prosodic models). Attentional-prosodic models manifest superior performance to Attentional ones for the task of word-learning. The Attentional-prosodic DQN outperforms existing word-learning models for the same task. PMID:29441027

Racial bias shapes social reinforcement learning.

PubMed

Lindström, Björn; Selbing, Ida; Molapour, Tanaz; Olsson, Andreas

2014-03-01

Both emotional facial expressions and markers of racial-group belonging are ubiquitous signals in social interaction, but little is known about how these signals together affect future behavior through learning. To address this issue, we investigated how emotional (threatening or friendly) in-group and out-group faces reinforced behavior in a reinforcement-learning task. We asked whether reinforcement learning would be modulated by intergroup attitudes (i.e., racial bias). The results showed that individual differences in racial bias critically modulated reinforcement learning. As predicted, racial bias was associated with more efficiently learned avoidance of threatening out-group individuals. We used computational modeling analysis to quantitatively delimit the underlying processes affected by social reinforcement. These analyses showed that racial bias modulates the rate at which exposure to threatening out-group individuals is transformed into future avoidance behavior. In concert, these results shed new light on the learning processes underlying social interaction with racial-in-group and out-group individuals.

Multi-agent Reinforcement Learning Model for Effective Action Selection

NASA Astrophysics Data System (ADS)

Youk, Sang Jo; Lee, Bong Keun

Reinforcement learning is a sub area of machine learning concerned with how an agent ought to take actions in an environment so as to maximize some notion of long-term reward. In the case of multi-agent, especially, which state space and action space gets very enormous in compared to single agent, so it needs to take most effective measure available select the action strategy for effective reinforcement learning. This paper proposes a multi-agent reinforcement learning model based on fuzzy inference system in order to improve learning collect speed and select an effective action in multi-agent. This paper verifies an effective action select strategy through evaluation tests based on Robocop Keep away which is one of useful test-beds for multi-agent. Our proposed model can apply to evaluate efficiency of the various intelligent multi-agents and also can apply to strategy and tactics of robot soccer system.

Simultaneous vibration control and energy harvesting using actor-critic based reinforcement learning

NASA Astrophysics Data System (ADS)

Loong, Cheng Ning; Chang, C. C.; Dimitrakopoulos, Elias G.

2018-03-01

Mitigating excessive vibration of civil engineering structures using various types of devices has been a conspicuous research topic in the past few decades. Some devices, such as electromagnetic transducers, which have a capability of exerting control forces while simultaneously harvesting energy, have been proposed recently. These devices make possible a self-regenerative system that can semi-actively mitigate structural vibration without the need of external energy. Integrating mechanical, electrical components, and control algorithms, these devices open up a new research domain that needs to be addressed. In this study, the feasibility of using an actor-critic based reinforcement learning control algorithm for simultaneous vibration control and energy harvesting for a civil engineering structure is investigated. The actor-critic based reinforcement learning control algorithm is a real-time, model-free adaptive technique that can adjust the controller parameters based on observations and reward signals without knowing the system characteristics. It is suitable for the control of a partially known nonlinear system with uncertain parameters. The feasibility of implementing this algorithm on a building structure equipped with an electromagnetic damper will be investigated in this study. Issues related to the modelling of learning algorithm, initialization and convergence will be presented and discussed.

Probabilistic Reinforcement Learning in Adults with Autism Spectrum Disorders

PubMed Central

Solomon, Marjorie; Smith, Anne C.; Frank, Michael J.; Ly, Stanford; Carter, Cameron S.

2017-01-01

Background Autism spectrum disorders (ASDs) can be conceptualized as disorders of learning, however there have been few experimental studies taking this perspective. Methods We examined the probabilistic reinforcement learning performance of 28 adults with ASDs and 30 typically developing adults on a task requiring learning relationships between three stimulus pairs consisting of Japanese characters with feedback that was valid with different probabilities (80%, 70%, and 60%). Both univariate and Bayesian state–space data analytic methods were employed. Hypotheses were based on the extant literature as well as on neurobiological and computational models of reinforcement learning. Results Both groups learned the task after training. However, there were group differences in early learning in the first task block where individuals with ASDs acquired the most frequently accurately reinforced stimulus pair (80%) comparably to typically developing individuals; exhibited poorer acquisition of the less frequently reinforced 70% pair as assessed by state–space learning curves; and outperformed typically developing individuals on the near chance (60%) pair. Individuals with ASDs also demonstrated deficits in using positive feedback to exploit rewarded choices. Conclusions Results support the contention that individuals with ASDs are slower learners. Based on neurobiology and on the results of computational modeling, one interpretation of this pattern of findings is that impairments are related to deficits in flexible updating of reinforcement history as mediated by the orbito-frontal cortex, with spared functioning of the basal ganglia. This hypothesis about the pathophysiology of learning in ASDs can be tested using functional magnetic resonance imaging. PMID:21425243

Model-based hierarchical reinforcement learning and human action control

PubMed Central

Botvinick, Matthew; Weinstein, Ari

2014-01-01

Recent work has reawakened interest in goal-directed or ‘model-based’ choice, where decisions are based on prospective evaluation of potential action outcomes. Concurrently, there has been growing attention to the role of hierarchy in decision-making and action control. We focus here on the intersection between these two areas of interest, considering the topic of hierarchical model-based control. To characterize this form of action control, we draw on the computational framework of hierarchical reinforcement learning, using this to interpret recent empirical findings. The resulting picture reveals how hierarchical model-based mechanisms might play a special and pivotal role in human decision-making, dramatically extending the scope and complexity of human behaviour. PMID:25267822

Rational and Mechanistic Perspectives on Reinforcement Learning

ERIC Educational Resources Information Center

Chater, Nick

2009-01-01

This special issue describes important recent developments in applying reinforcement learning models to capture neural and cognitive function. But reinforcement learning, as a theoretical framework, can apply at two very different levels of description: "mechanistic" and "rational." Reinforcement learning is often viewed in mechanistic terms--as…

Effect of reinforcement learning on coordination of multiangent systems

NASA Astrophysics Data System (ADS)

Bukkapatnam, Satish T. S.; Gao, Greg

2000-12-01

For effective coordination of distributed environments involving multiagent systems, learning ability of each agent in the environment plays a crucial role. In this paper, we develop a simple group learning method based on reinforcement, and study its effect on coordination through application to a supply chain procurement scenario involving a computer manufacturer. Here, all parties are represented by self-interested, autonomous agents, each capable of performing specific simple tasks. They negotiate with each other to perform complex tasks and thus coordinate supply chain procurement. Reinforcement learning is intended to enable each agent to reach a best negotiable price within a shortest possible time. Our simulations of the application scenario under different learning strategies reveals the positive effects of reinforcement learning on an agent's as well as the system's performance.

Reinforcement and inference in cross-situational word learning.

PubMed

Tilles, Paulo F C; Fontanari, José F

2013-01-01

Cross-situational word learning is based on the notion that a learner can determine the referent of a word by finding something in common across many observed uses of that word. Here we propose an adaptive learning algorithm that contains a parameter that controls the strength of the reinforcement applied to associations between concurrent words and referents, and a parameter that regulates inference, which includes built-in biases, such as mutual exclusivity, and information of past learning events. By adjusting these parameters so that the model predictions agree with data from representative experiments on cross-situational word learning, we were able to explain the learning strategies adopted by the participants of those experiments in terms of a trade-off between reinforcement and inference. These strategies can vary wildly depending on the conditions of the experiments. For instance, for fast mapping experiments (i.e., the correct referent could, in principle, be inferred in a single observation) inference is prevalent, whereas for segregated contextual diversity experiments (i.e., the referents are separated in groups and are exhibited with members of their groups only) reinforcement is predominant. Other experiments are explained with more balanced doses of reinforcement and inference.

Batch Mode Reinforcement Learning based on the Synthesis of Artificial Trajectories

PubMed Central

Fonteneau, Raphael; Murphy, Susan A.; Wehenkel, Louis; Ernst, Damien

2013-01-01

In this paper, we consider the batch mode reinforcement learning setting, where the central problem is to learn from a sample of trajectories a policy that satisfies or optimizes a performance criterion. We focus on the continuous state space case for which usual resolution schemes rely on function approximators either to represent the underlying control problem or to represent its value function. As an alternative to the use of function approximators, we rely on the synthesis of “artificial trajectories” from the given sample of trajectories, and show that this idea opens new avenues for designing and analyzing algorithms for batch mode reinforcement learning. PMID:24049244

Bio-robots automatic navigation with graded electric reward stimulation based on Reinforcement Learning.

PubMed

Zhang, Chen; Sun, Chao; Gao, Liqiang; Zheng, Nenggan; Chen, Weidong; Zheng, Xiaoxiang

2013-01-01

Bio-robots based on brain computer interface (BCI) suffer from the lack of considering the characteristic of the animals in navigation. This paper proposed a new method for bio-robots' automatic navigation combining the reward generating algorithm base on Reinforcement Learning (RL) with the learning intelligence of animals together. Given the graded electrical reward, the animal e.g. the rat, intends to seek the maximum reward while exploring an unknown environment. Since the rat has excellent spatial recognition, the rat-robot and the RL algorithm can convergent to an optimal route by co-learning. This work has significant inspiration for the practical development of bio-robots' navigation with hybrid intelligence.

Partial Planning Reinforcement Learning

DTIC Science & Technology

2012-08-31

Research Office P.O. Box 12211 Research Triangle Park, NC 27709-2211 15. SUBJECT TERMS Reinforcement Learning, Bayesian Optimization, Active ... Learning , Action Model Learning, Decision Theoretic Assistance Prasad Tadepalli, Alan Fern Oregon State University Office of Sponsored Programs Oregon State

Collaborating Fuzzy Reinforcement Learning Agents

NASA Technical Reports Server (NTRS)

Berenji, Hamid R.

1997-01-01

Earlier, we introduced GARIC-Q, a new method for doing incremental Dynamic Programming using a society of intelligent agents which are controlled at the top level by Fuzzy Relearning and at the local level, each agent learns and operates based on ANTARCTIC, a technique for fuzzy reinforcement learning. In this paper, we show that it is possible for these agents to compete in order to affect the selected control policy but at the same time, they can collaborate while investigating the state space. In this model, the evaluator or the critic learns by observing all the agents behaviors but the control policy changes only based on the behavior of the winning agent also known as the super agent.

Structure identification in fuzzy inference using reinforcement learning

NASA Technical Reports Server (NTRS)

Berenji, Hamid R.; Khedkar, Pratap

1993-01-01

In our previous work on the GARIC architecture, we have shown that the system can start with surface structure of the knowledge base (i.e., the linguistic expression of the rules) and learn the deep structure (i.e., the fuzzy membership functions of the labels used in the rules) by using reinforcement learning. Assuming the surface structure, GARIC refines the fuzzy membership functions used in the consequents of the rules using a gradient descent procedure. This hybrid fuzzy logic and reinforcement learning approach can learn to balance a cart-pole system and to backup a truck to its docking location after a few trials. In this paper, we discuss how to do structure identification using reinforcement learning in fuzzy inference systems. This involves identifying both surface as well as deep structure of the knowledge base. The term set of fuzzy linguistic labels used in describing the values of each control variable must be derived. In this process, splitting a label refers to creating new labels which are more granular than the original label and merging two labels creates a more general label. Splitting and merging of labels directly transform the structure of the action selection network used in GARIC by increasing or decreasing the number of hidden layer nodes.

Negative reinforcement learning is affected in substance dependence.

PubMed

Thompson, Laetitia L; Claus, Eric D; Mikulich-Gilbertson, Susan K; Banich, Marie T; Crowley, Thomas; Krmpotich, Theodore; Miller, David; Tanabe, Jody

2012-06-01

Negative reinforcement results in behavior to escape or avoid an aversive outcome. Withdrawal symptoms are purported to be negative reinforcers in perpetuating substance dependence, but little is known about negative reinforcement learning in this population. The purpose of this study was to examine reinforcement learning in substance dependent individuals (SDI), with an emphasis on assessing negative reinforcement learning. We modified the Iowa Gambling Task to separately assess positive and negative reinforcement. We hypothesized that SDI would show differences in negative reinforcement learning compared to controls and we investigated whether learning differed as a function of the relative magnitude or frequency of the reinforcer. Thirty subjects dependent on psychostimulants were compared with 28 community controls on a decision making task that manipulated outcome frequencies and magnitudes and required an action to avoid a negative outcome. SDI did not learn to avoid negative outcomes to the same degree as controls. This difference was driven by the magnitude, not the frequency, of negative feedback. In contrast, approach behaviors in response to positive reinforcement were similar in both groups. Our findings are consistent with a specific deficit in negative reinforcement learning in SDI. SDI were relatively insensitive to the magnitude, not frequency, of loss. If this generalizes to drug-related stimuli, it suggests that repeated episodes of withdrawal may drive relapse more than the severity of a single episode. Copyright © 2011 Elsevier Ireland Ltd. All rights reserved.

Learning and tuning fuzzy logic controllers through reinforcements.

PubMed

Berenji, H R; Khedkar, P

1992-01-01

A method for learning and tuning a fuzzy logic controller based on reinforcements from a dynamic system is presented. It is shown that: the generalized approximate-reasoning-based intelligent control (GARIC) architecture learns and tunes a fuzzy logic controller even when only weak reinforcement, such as a binary failure signal, is available; introduces a new conjunction operator in computing the rule strengths of fuzzy control rules; introduces a new localized mean of maximum (LMOM) method in combining the conclusions of several firing control rules; and learns to produce real-valued control actions. Learning is achieved by integrating fuzzy inference into a feedforward network, which can then adaptively improve performance by using gradient descent methods. The GARIC architecture is applied to a cart-pole balancing system and demonstrates significant improvements in terms of the speed of learning and robustness to changes in the dynamic system's parameters over previous schemes for cart-pole balancing.

Multi-Objective Reinforcement Learning for Cognitive Radio-Based Satellite Communications

NASA Technical Reports Server (NTRS)

Ferreira, Paulo Victor R.; Paffenroth, Randy; Wyglinski, Alexander M.; Hackett, Timothy M.; Bilen, Sven G.; Reinhart, Richard C.; Mortensen, Dale J.

2016-01-01

Previous research on cognitive radios has addressed the performance of various machine-learning and optimization techniques for decision making of terrestrial link properties. In this paper, we present our recent investigations with respect to reinforcement learning that potentially can be employed by future cognitive radios installed onboard satellite communications systems specifically tasked with radio resource management. This work analyzes the performance of learning, reasoning, and decision making while considering multiple objectives for time-varying communications channels, as well as different cross-layer requirements. Based on the urgent demand for increased bandwidth, which is being addressed by the next generation of high-throughput satellites, the performance of cognitive radio is assessed considering links between a geostationary satellite and a fixed ground station operating at Ka-band (26 GHz). Simulation results show multiple objective performance improvements of more than 3.5 times for clear sky conditions and 6.8 times for rain conditions.

Multi-Objective Reinforcement Learning for Cognitive Radio Based Satellite Communications

NASA Technical Reports Server (NTRS)

Ferreira, Paulo; Paffenroth, Randy; Wyglinski, Alexander; Hackett, Timothy; Bilen, Sven; Reinhart, Richard; Mortensen, Dale John

2016-01-01

Previous research on cognitive radios has addressed the performance of various machine learning and optimization techniques for decision making of terrestrial link properties. In this paper, we present our recent investigations with respect to reinforcement learning that potentially can be employed by future cognitive radios installed onboard satellite communications systems specifically tasked with radio resource management. This work analyzes the performance of learning, reasoning, and decision making while considering multiple objectives for time-varying communications channels, as well as different crosslayer requirements. Based on the urgent demand for increased bandwidth, which is being addressed by the next generation of high-throughput satellites, the performance of cognitive radio is assessed considering links between a geostationary satellite and a fixed ground station operating at Ka-band (26 GHz). Simulation results show multiple objective performance improvements of more than 3:5 times for clear sky conditions and 6:8 times for rain conditions.

Improving Robot Motor Learning with Negatively Valenced Reinforcement Signals

PubMed Central

Navarro-Guerrero, Nicolás; Lowe, Robert J.; Wermter, Stefan

2017-01-01

Both nociception and punishment signals have been used in robotics. However, the potential for using these negatively valenced types of reinforcement learning signals for robot learning has not been exploited in detail yet. Nociceptive signals are primarily used as triggers of preprogrammed action sequences. Punishment signals are typically disembodied, i.e., with no or little relation to the agent-intrinsic limitations, and they are often used to impose behavioral constraints. Here, we provide an alternative approach for nociceptive signals as drivers of learning rather than simple triggers of preprogrammed behavior. Explicitly, we use nociception to expand the state space while we use punishment as a negative reinforcement learning signal. We compare the performance—in terms of task error, the amount of perceived nociception, and length of learned action sequences—of different neural networks imbued with punishment-based reinforcement signals for inverse kinematic learning. We contrast the performance of a version of the neural network that receives nociceptive inputs to that without such a process. Furthermore, we provide evidence that nociception can improve learning—making the algorithm more robust against network initializations—as well as behavioral performance by reducing the task error, perceived nociception, and length of learned action sequences. Moreover, we provide evidence that punishment, at least as typically used within reinforcement learning applications, may be detrimental in all relevant metrics. PMID:28420976

From creatures of habit to goal-directed learners: Tracking the developmental emergence of model-based reinforcement learning

PubMed Central

Decker, Johannes H.; Otto, A. Ross; Daw, Nathaniel D.; Hartley, Catherine A.

2016-01-01

Theoretical models distinguish two decision-making strategies that have been formalized in reinforcement-learning theory. A model-based strategy leverages a cognitive model of potential actions and their consequences to make goal-directed choices, whereas a model-free strategy evaluates actions based solely on their reward history. Research in adults has begun to elucidate the psychological mechanisms and neural substrates underlying these learning processes and factors that influence their relative recruitment. However, the developmental trajectory of these evaluative strategies has not been well characterized. In this study, children, adolescents, and adults, performed a sequential reinforcement-learning task that enables estimation of model-based and model-free contributions to choice. Whereas a model-free strategy was evident in choice behavior across all age groups, evidence of a model-based strategy only emerged during adolescence and continued to increase into adulthood. These results suggest that recruitment of model-based valuation systems represents a critical cognitive component underlying the gradual maturation of goal-directed behavior. PMID:27084852

From Creatures of Habit to Goal-Directed Learners: Tracking the Developmental Emergence of Model-Based Reinforcement Learning.

PubMed

Decker, Johannes H; Otto, A Ross; Daw, Nathaniel D; Hartley, Catherine A

2016-06-01

Theoretical models distinguish two decision-making strategies that have been formalized in reinforcement-learning theory. A model-based strategy leverages a cognitive model of potential actions and their consequences to make goal-directed choices, whereas a model-free strategy evaluates actions based solely on their reward history. Research in adults has begun to elucidate the psychological mechanisms and neural substrates underlying these learning processes and factors that influence their relative recruitment. However, the developmental trajectory of these evaluative strategies has not been well characterized. In this study, children, adolescents, and adults performed a sequential reinforcement-learning task that enabled estimation of model-based and model-free contributions to choice. Whereas a model-free strategy was apparent in choice behavior across all age groups, a model-based strategy was absent in children, became evident in adolescents, and strengthened in adults. These results suggest that recruitment of model-based valuation systems represents a critical cognitive component underlying the gradual maturation of goal-directed behavior. © The Author(s) 2016.

Reinforcement Learning Based Web Service Compositions for Mobile Business

NASA Astrophysics Data System (ADS)

Zhou, Juan; Chen, Shouming

In this paper, we propose a new solution to Reactive Web Service Composition, via molding with Reinforcement Learning, and introducing modified (alterable) QoS variables into the model as elements in the Markov Decision Process tuple. Moreover, we give an example of Reactive-WSC-based mobile banking, to demonstrate the intrinsic capability of the solution in question of obtaining the optimized service composition, characterized by (alterable) target QoS variable sets with optimized values. Consequently, we come to the conclusion that the solution has decent potentials in boosting customer experiences and qualities of services in Web Services, and those in applications in the whole electronic commerce and business sector.

Stress enhances model-free reinforcement learning only after negative outcome.

PubMed

Park, Heyeon; Lee, Daeyeol; Chey, Jeanyung

2017-01-01

Previous studies found that stress shifts behavioral control by promoting habits while decreasing goal-directed behaviors during reward-based decision-making. It is, however, unclear how stress disrupts the relative contribution of the two systems controlling reward-seeking behavior, i.e. model-free (or habit) and model-based (or goal-directed). Here, we investigated whether stress biases the contribution of model-free and model-based reinforcement learning processes differently depending on the valence of outcome, and whether stress alters the learning rate, i.e., how quickly information from the new environment is incorporated into choices. Participants were randomly assigned to either a stress or a control condition, and performed a two-stage Markov decision-making task in which the reward probabilities underwent periodic reversals without notice. We found that stress increased the contribution of model-free reinforcement learning only after negative outcome. Furthermore, stress decreased the learning rate. The results suggest that stress diminishes one's ability to make adaptive choices in multiple aspects of reinforcement learning. This finding has implications for understanding how stress facilitates maladaptive habits, such as addictive behavior, and other dysfunctional behaviors associated with stress in clinical and educational contexts.

Stress enhances model-free reinforcement learning only after negative outcome

PubMed Central

Lee, Daeyeol

2017-01-01

Previous studies found that stress shifts behavioral control by promoting habits while decreasing goal-directed behaviors during reward-based decision-making. It is, however, unclear how stress disrupts the relative contribution of the two systems controlling reward-seeking behavior, i.e. model-free (or habit) and model-based (or goal-directed). Here, we investigated whether stress biases the contribution of model-free and model-based reinforcement learning processes differently depending on the valence of outcome, and whether stress alters the learning rate, i.e., how quickly information from the new environment is incorporated into choices. Participants were randomly assigned to either a stress or a control condition, and performed a two-stage Markov decision-making task in which the reward probabilities underwent periodic reversals without notice. We found that stress increased the contribution of model-free reinforcement learning only after negative outcome. Furthermore, stress decreased the learning rate. The results suggest that stress diminishes one’s ability to make adaptive choices in multiple aspects of reinforcement learning. This finding has implications for understanding how stress facilitates maladaptive habits, such as addictive behavior, and other dysfunctional behaviors associated with stress in clinical and educational contexts. PMID:28723943

Reinforcement learning state estimator.

PubMed

Morimoto, Jun; Doya, Kenji

2007-03-01

In this study, we propose a novel use of reinforcement learning for estimating hidden variables and parameters of nonlinear dynamical systems. A critical issue in hidden-state estimation is that we cannot directly observe estimation errors. However, by defining errors of observable variables as a delayed penalty, we can apply a reinforcement learning frame-work to state estimation problems. Specifically, we derive a method to construct a nonlinear state estimator by finding an appropriate feedback input gain using the policy gradient method. We tested the proposed method on single pendulum dynamics and show that the joint angle variable could be successfully estimated by observing only the angular velocity, and vice versa. In addition, we show that we could acquire a state estimator for the pendulum swing-up task in which a swing-up controller is also acquired by reinforcement learning simultaneously. Furthermore, we demonstrate that it is possible to estimate the dynamics of the pendulum itself while the hidden variables are estimated in the pendulum swing-up task. Application of the proposed method to a two-linked biped model is also presented.

Stress affects instrumental learning based on positive or negative reinforcement in interaction with personality in domestic horses

PubMed Central

Valenchon, Mathilde; Lévy, Frédéric; Moussu, Chantal; Lansade, Léa

2017-01-01

The present study investigated how stress affects instrumental learning performance in horses (Equus caballus) depending on the type of reinforcement. Horses were assigned to four groups (N = 15 per group); each group received training with negative or positive reinforcement in the presence or absence of stressors unrelated to the learning task. The instrumental learning task consisted of the horse entering one of two compartments at the appearance of a visual signal given by the experimenter. In the absence of stressors unrelated to the task, learning performance did not differ between negative and positive reinforcements. The presence of stressors unrelated to the task (exposure to novel and sudden stimuli) impaired learning performance. Interestingly, this learning deficit was smaller when the negative reinforcement was used. The negative reinforcement, considered as a stressor related to the task, could have counterbalanced the impact of the extrinsic stressor by focusing attention toward the learning task. In addition, learning performance appears to differ between certain dimensions of personality depending on the presence of stressors and the type of reinforcement. These results suggest that when negative reinforcement is used (i.e. stressor related to the task), the most fearful horses may be the best performers in the absence of stressors but the worst performers when stressors are present. On the contrary, when positive reinforcement is used, the most fearful horses appear to be consistently the worst performers, with and without exposure to stressors unrelated to the learning task. This study is the first to demonstrate in ungulates that stress affects learning performance differentially according to the type of reinforcement and in interaction with personality. It provides fundamental and applied perspectives in the understanding of the relationships between personality and training abilities. PMID:28475581

Predictive representations can link model-based reinforcement learning to model-free mechanisms.

PubMed

Russek, Evan M; Momennejad, Ida; Botvinick, Matthew M; Gershman, Samuel J; Daw, Nathaniel D

2017-09-01

Humans and animals are capable of evaluating actions by considering their long-run future rewards through a process described using model-based reinforcement learning (RL) algorithms. The mechanisms by which neural circuits perform the computations prescribed by model-based RL remain largely unknown; however, multiple lines of evidence suggest that neural circuits supporting model-based behavior are structurally homologous to and overlapping with those thought to carry out model-free temporal difference (TD) learning. Here, we lay out a family of approaches by which model-based computation may be built upon a core of TD learning. The foundation of this framework is the successor representation, a predictive state representation that, when combined with TD learning of value predictions, can produce a subset of the behaviors associated with model-based learning, while requiring less decision-time computation than dynamic programming. Using simulations, we delineate the precise behavioral capabilities enabled by evaluating actions using this approach, and compare them to those demonstrated by biological organisms. We then introduce two new algorithms that build upon the successor representation while progressively mitigating its limitations. Because this framework can account for the full range of observed putatively model-based behaviors while still utilizing a core TD framework, we suggest that it represents a neurally plausible family of mechanisms for model-based evaluation.

Predictive representations can link model-based reinforcement learning to model-free mechanisms

PubMed Central

Botvinick, Matthew M.

2017-01-01

Humans and animals are capable of evaluating actions by considering their long-run future rewards through a process described using model-based reinforcement learning (RL) algorithms. The mechanisms by which neural circuits perform the computations prescribed by model-based RL remain largely unknown; however, multiple lines of evidence suggest that neural circuits supporting model-based behavior are structurally homologous to and overlapping with those thought to carry out model-free temporal difference (TD) learning. Here, we lay out a family of approaches by which model-based computation may be built upon a core of TD learning. The foundation of this framework is the successor representation, a predictive state representation that, when combined with TD learning of value predictions, can produce a subset of the behaviors associated with model-based learning, while requiring less decision-time computation than dynamic programming. Using simulations, we delineate the precise behavioral capabilities enabled by evaluating actions using this approach, and compare them to those demonstrated by biological organisms. We then introduce two new algorithms that build upon the successor representation while progressively mitigating its limitations. Because this framework can account for the full range of observed putatively model-based behaviors while still utilizing a core TD framework, we suggest that it represents a neurally plausible family of mechanisms for model-based evaluation. PMID:28945743

Learning and tuning fuzzy logic controllers through reinforcements

NASA Technical Reports Server (NTRS)

Berenji, Hamid R.; Khedkar, Pratap

1992-01-01

A new method for learning and tuning a fuzzy logic controller based on reinforcements from a dynamic system is presented. In particular, our Generalized Approximate Reasoning-based Intelligent Control (GARIC) architecture: (1) learns and tunes a fuzzy logic controller even when only weak reinforcements, such as a binary failure signal, is available; (2) introduces a new conjunction operator in computing the rule strengths of fuzzy control rules; (3) introduces a new localized mean of maximum (LMOM) method in combining the conclusions of several firing control rules; and (4) learns to produce real-valued control actions. Learning is achieved by integrating fuzzy inference into a feedforward network, which can then adaptively improve performance by using gradient descent methods. We extend the AHC algorithm of Barto, Sutton, and Anderson to include the prior control knowledge of human operators. The GARIC architecture is applied to a cart-pole balancing system and has demonstrated significant improvements in terms of the speed of learning and robustness to changes in the dynamic system's parameters over previous schemes for cart-pole balancing.

Learning and tuning fuzzy logic controllers through reinforcements

NASA Technical Reports Server (NTRS)

Berenji, Hamid R.; Khedkar, Pratap

1992-01-01

This paper presents a new method for learning and tuning a fuzzy logic controller based on reinforcements from a dynamic system. In particular, our generalized approximate reasoning-based intelligent control (GARIC) architecture (1) learns and tunes a fuzzy logic controller even when only weak reinforcement, such as a binary failure signal, is available; (2) introduces a new conjunction operator in computing the rule strengths of fuzzy control rules; (3) introduces a new localized mean of maximum (LMOM) method in combining the conclusions of several firing control rules; and (4) learns to produce real-valued control actions. Learning is achieved by integrating fuzzy inference into a feedforward neural network, which can then adaptively improve performance by using gradient descent methods. We extend the AHC algorithm of Barto et al. (1983) to include the prior control knowledge of human operators. The GARIC architecture is applied to a cart-pole balancing system and demonstrates significant improvements in terms of the speed of learning and robustness to changes in the dynamic system's parameters over previous schemes for cart-pole balancing.

Reinforcement Learning with Orthonormal Basis Adaptation Based on Activity-Oriented Index Allocation

NASA Astrophysics Data System (ADS)

Satoh, Hideki

An orthonormal basis adaptation method for function approximation was developed and applied to reinforcement learning with multi-dimensional continuous state space. First, a basis used for linear function approximation of a control function is set to an orthonormal basis. Next, basis elements with small activities are replaced with other candidate elements as learning progresses. As this replacement is repeated, the number of basis elements with large activities increases. Example chaos control problems for multiple logistic maps were solved, demonstrating that the method for adapting an orthonormal basis can modify a basis while holding the orthonormality in accordance with changes in the environment to improve the performance of reinforcement learning and to eliminate the adverse effects of redundant noisy states.

Separation of time-based and trial-based accounts of the partial reinforcement extinction effect.

PubMed

Bouton, Mark E; Woods, Amanda M; Todd, Travis P

2014-01-01

Two appetitive conditioning experiments with rats examined time-based and trial-based accounts of the partial reinforcement extinction effect (PREE). In the PREE, the loss of responding that occurs in extinction is slower when the conditioned stimulus (CS) has been paired with a reinforcer on some of its presentations (partially reinforced) instead of every presentation (continuously reinforced). According to a time-based or "time-accumulation" view (e.g., Gallistel and Gibbon, 2000), the PREE occurs because the organism has learned in partial reinforcement to expect the reinforcer after a larger amount of time has accumulated in the CS over trials. In contrast, according to a trial-based view (e.g., Capaldi, 1967), the PREE occurs because the organism has learned in partial reinforcement to expect the reinforcer after a larger number of CS presentations. Experiment 1 used a procedure that equated partially and continuously reinforced groups on their expected times to reinforcement during conditioning. A PREE was still observed. Experiment 2 then used an extinction procedure that allowed time in the CS and the number of trials to accumulate differentially through extinction. The PREE was still evident when responding was examined as a function of expected time units to the reinforcer, but was eliminated when responding was examined as a function of expected trial units to the reinforcer. There was no evidence that the animal responded according to the ratio of time accumulated during the CS in extinction over the time in the CS expected before the reinforcer. The results thus favor a trial-based account over a time-based account of extinction and the PREE. This article is part of a Special Issue entitled: Associative and Temporal Learning. Copyright © 2013 Elsevier B.V. All rights reserved.

Enhanced Experience Replay for Deep Reinforcement Learning

DTIC Science & Technology

2015-11-01

ARL-TR-7538 ● NOV 2015 US Army Research Laboratory Enhanced Experience Replay for Deep Reinforcement Learning by David Doria...Experience Replay for Deep Reinforcement Learning by David Doria, Bryan Dawson, and Manuel Vindiola Computational and Information Sciences Directorate...

Application of fuzzy logic-neural network based reinforcement learning to proximity and docking operations: Translational controller results

NASA Technical Reports Server (NTRS)

Jani, Yashvant

1992-01-01

The reinforcement learning techniques developed at Ames Research Center are being applied to proximity and docking operations using the Shuttle and Solar Maximum Mission (SMM) satellite simulation. In utilizing these fuzzy learning techniques, we also use the Approximate Reasoning based Intelligent Control (ARIC) architecture, and so we use two terms interchangeable to imply the same. This activity is carried out in the Software Technology Laboratory utilizing the Orbital Operations Simulator (OOS). This report is the deliverable D3 in our project activity and provides the test results of the fuzzy learning translational controller. This report is organized in six sections. Based on our experience and analysis with the attitude controller, we have modified the basic configuration of the reinforcement learning algorithm in ARIC as described in section 2. The shuttle translational controller and its implementation in fuzzy learning architecture is described in section 3. Two test cases that we have performed are described in section 4. Our results and conclusions are discussed in section 5, and section 6 provides future plans and summary for the project.

Learning to reach by reinforcement learning using a receptive field based function approximation approach with continuous actions.

PubMed

Tamosiunaite, Minija; Asfour, Tamim; Wörgötter, Florentin

2009-03-01

Reinforcement learning methods can be used in robotics applications especially for specific target-oriented problems, for example the reward-based recalibration of goal directed actions. To this end still relatively large and continuous state-action spaces need to be efficiently handled. The goal of this paper is, thus, to develop a novel, rather simple method which uses reinforcement learning with function approximation in conjunction with different reward-strategies for solving such problems. For the testing of our method, we use a four degree-of-freedom reaching problem in 3D-space simulated by a two-joint robot arm system with two DOF each. Function approximation is based on 4D, overlapping kernels (receptive fields) and the state-action space contains about 10,000 of these. Different types of reward structures are being compared, for example, reward-on- touching-only against reward-on-approach. Furthermore, forbidden joint configurations are punished. A continuous action space is used. In spite of a rather large number of states and the continuous action space these reward/punishment strategies allow the system to find a good solution usually within about 20 trials. The efficiency of our method demonstrated in this test scenario suggests that it might be possible to use it on a real robot for problems where mixed rewards can be defined in situations where other types of learning might be difficult.

The Curse of Planning: Dissecting multiple reinforcement learning systems by taxing the central executive

PubMed Central

Otto, A. Ross; Gershman, Samuel J.; Markman, Arthur B.; Daw, Nathaniel D.

2013-01-01

A number of accounts of human and animal behavior posit the operation of parallel and competing valuation systems in the control of choice behavior. Along these lines, a flexible but computationally expensive model-based reinforcement learning system has been contrasted with a less flexible but more efficient model-free reinforcement learning system. The factors governing which system controls behavior—and under what circumstances—are still unclear. Based on the hypothesis that model-based reinforcement learning requires cognitive resources, we demonstrate that having human decision-makers perform a demanding secondary task engenders increased reliance on a model-free reinforcement learning strategy. Further, we show that across trials, people negotiate this tradeoff dynamically as a function of concurrent executive function demands and their choice latencies reflect the computational expenses of the strategy employed. These results demonstrate that competition between multiple learning systems can be controlled on a trial-by-trial basis by modulating the availability of cognitive resources. PMID:23558545

Human-level control through deep reinforcement learning.

PubMed

Mnih, Volodymyr; Kavukcuoglu, Koray; Silver, David; Rusu, Andrei A; Veness, Joel; Bellemare, Marc G; Graves, Alex; Riedmiller, Martin; Fidjeland, Andreas K; Ostrovski, Georg; Petersen, Stig; Beattie, Charles; Sadik, Amir; Antonoglou, Ioannis; King, Helen; Kumaran, Dharshan; Wierstra, Daan; Legg, Shane; Hassabis, Demis

2015-02-26

The theory of reinforcement learning provides a normative account, deeply rooted in psychological and neuroscientific perspectives on animal behaviour, of how agents may optimize their control of an environment. To use reinforcement learning successfully in situations approaching real-world complexity, however, agents are confronted with a difficult task: they must derive efficient representations of the environment from high-dimensional sensory inputs, and use these to generalize past experience to new situations. Remarkably, humans and other animals seem to solve this problem through a harmonious combination of reinforcement learning and hierarchical sensory processing systems, the former evidenced by a wealth of neural data revealing notable parallels between the phasic signals emitted by dopaminergic neurons and temporal difference reinforcement learning algorithms. While reinforcement learning agents have achieved some successes in a variety of domains, their applicability has previously been limited to domains in which useful features can be handcrafted, or to domains with fully observed, low-dimensional state spaces. Here we use recent advances in training deep neural networks to develop a novel artificial agent, termed a deep Q-network, that can learn successful policies directly from high-dimensional sensory inputs using end-to-end reinforcement learning. We tested this agent on the challenging domain of classic Atari 2600 games. We demonstrate that the deep Q-network agent, receiving only the pixels and the game score as inputs, was able to surpass the performance of all previous algorithms and achieve a level comparable to that of a professional human games tester across a set of 49 games, using the same algorithm, network architecture and hyperparameters. This work bridges the divide between high-dimensional sensory inputs and actions, resulting in the first artificial agent that is capable of learning to excel at a diverse array of challenging tasks.

Human-level control through deep reinforcement learning

NASA Astrophysics Data System (ADS)

Mnih, Volodymyr; Kavukcuoglu, Koray; Silver, David; Rusu, Andrei A.; Veness, Joel; Bellemare, Marc G.; Graves, Alex; Riedmiller, Martin; Fidjeland, Andreas K.; Ostrovski, Georg; Petersen, Stig; Beattie, Charles; Sadik, Amir; Antonoglou, Ioannis; King, Helen; Kumaran, Dharshan; Wierstra, Daan; Legg, Shane; Hassabis, Demis

2015-02-01

The theory of reinforcement learning provides a normative account, deeply rooted in psychological and neuroscientific perspectives on animal behaviour, of how agents may optimize their control of an environment. To use reinforcement learning successfully in situations approaching real-world complexity, however, agents are confronted with a difficult task: they must derive efficient representations of the environment from high-dimensional sensory inputs, and use these to generalize past experience to new situations. Remarkably, humans and other animals seem to solve this problem through a harmonious combination of reinforcement learning and hierarchical sensory processing systems, the former evidenced by a wealth of neural data revealing notable parallels between the phasic signals emitted by dopaminergic neurons and temporal difference reinforcement learning algorithms. While reinforcement learning agents have achieved some successes in a variety of domains, their applicability has previously been limited to domains in which useful features can be handcrafted, or to domains with fully observed, low-dimensional state spaces. Here we use recent advances in training deep neural networks to develop a novel artificial agent, termed a deep Q-network, that can learn successful policies directly from high-dimensional sensory inputs using end-to-end reinforcement learning. We tested this agent on the challenging domain of classic Atari 2600 games. We demonstrate that the deep Q-network agent, receiving only the pixels and the game score as inputs, was able to surpass the performance of all previous algorithms and achieve a level comparable to that of a professional human games tester across a set of 49 games, using the same algorithm, network architecture and hyperparameters. This work bridges the divide between high-dimensional sensory inputs and actions, resulting in the first artificial agent that is capable of learning to excel at a diverse array of challenging tasks.

Reinforcement learning signals in the human striatum distinguish learners from nonlearners during reward-based decision making.

PubMed

Schönberg, Tom; Daw, Nathaniel D; Joel, Daphna; O'Doherty, John P

2007-11-21

The computational framework of reinforcement learning has been used to forward our understanding of the neural mechanisms underlying reward learning and decision-making behavior. It is known that humans vary widely in their performance in decision-making tasks. Here, we used a simple four-armed bandit task in which subjects are almost evenly split into two groups on the basis of their performance: those who do learn to favor choice of the optimal action and those who do not. Using models of reinforcement learning we sought to determine the neural basis of these intrinsic differences in performance by scanning both groups with functional magnetic resonance imaging. We scanned 29 subjects while they performed the reward-based decision-making task. Our results suggest that these two groups differ markedly in the degree to which reinforcement learning signals in the striatum are engaged during task performance. While the learners showed robust prediction error signals in both the ventral and dorsal striatum during learning, the nonlearner group showed a marked absence of such signals. Moreover, the magnitude of prediction error signals in a region of dorsal striatum correlated significantly with a measure of behavioral performance across all subjects. These findings support a crucial role of prediction error signals, likely originating from dopaminergic midbrain neurons, in enabling learning of action selection preferences on the basis of obtained rewards. Thus, spontaneously observed individual differences in decision making performance demonstrate the suggested dependence of this type of learning on the functional integrity of the dopaminergic striatal system in humans.

Altered Risk-Based Decision Making following Adolescent Alcohol Use Results from an Imbalance in Reinforcement Learning in Rats

PubMed Central

Hart, Andrew S.; Collins, Anne L.; Bernstein, Ilene L.; Phillips, Paul E. M.

2012-01-01

Alcohol use during adolescence has profound and enduring consequences on decision-making under risk. However, the fundamental psychological processes underlying these changes are unknown. Here, we show that alcohol use produces over-fast learning for better-than-expected, but not worse-than-expected, outcomes without altering subjective reward valuation. We constructed a simple reinforcement learning model to simulate altered decision making using behavioral parameters extracted from rats with a history of adolescent alcohol use. Remarkably, the learning imbalance alone was sufficient to simulate the divergence in choice behavior observed between these groups of animals. These findings identify a selective alteration in reinforcement learning following adolescent alcohol use that can account for a robust change in risk-based decision making persisting into later life. PMID:22615989

Framework for robot skill learning using reinforcement learning

NASA Astrophysics Data System (ADS)

Wei, Yingzi; Zhao, Mingyang

2003-09-01

Robot acquiring skill is a process similar to human skill learning. Reinforcement learning (RL) is an on-line actor critic method for a robot to develop its skill. The reinforcement function has become the critical component for its effect of evaluating the action and guiding the learning process. We present an augmented reward function that provides a new way for RL controller to incorporate prior knowledge and experience into the RL controller. Also, the difference form of augmented reward function is considered carefully. The additional reward beyond conventional reward will provide more heuristic information for RL. In this paper, we present a strategy for the task of complex skill learning. Automatic robot shaping policy is to dissolve the complex skill into a hierarchical learning process. The new form of value function is introduced to attain smooth motion switching swiftly. We present a formal, but practical, framework for robot skill learning and also illustrate with an example the utility of method for learning skilled robot control on line.

Quantum reinforcement learning.

PubMed

Dong, Daoyi; Chen, Chunlin; Li, Hanxiong; Tarn, Tzyh-Jong

2008-10-01

The key approaches for machine learning, particularly learning in unknown probabilistic environments, are new representations and computation mechanisms. In this paper, a novel quantum reinforcement learning (QRL) method is proposed by combining quantum theory and reinforcement learning (RL). Inspired by the state superposition principle and quantum parallelism, a framework of a value-updating algorithm is introduced. The state (action) in traditional RL is identified as the eigen state (eigen action) in QRL. The state (action) set can be represented with a quantum superposition state, and the eigen state (eigen action) can be obtained by randomly observing the simulated quantum state according to the collapse postulate of quantum measurement. The probability of the eigen action is determined by the probability amplitude, which is updated in parallel according to rewards. Some related characteristics of QRL such as convergence, optimality, and balancing between exploration and exploitation are also analyzed, which shows that this approach makes a good tradeoff between exploration and exploitation using the probability amplitude and can speedup learning through the quantum parallelism. To evaluate the performance and practicability of QRL, several simulated experiments are given, and the results demonstrate the effectiveness and superiority of the QRL algorithm for some complex problems. This paper is also an effective exploration on the application of quantum computation to artificial intelligence.

Evolution with Reinforcement Learning in Negotiation

PubMed Central

Zou, Yi; Zhan, Wenjie; Shao, Yuan

2014-01-01

Adaptive behavior depends less on the details of the negotiation process and makes more robust predictions in the long term as compared to in the short term. However, the extant literature on population dynamics for behavior adjustment has only examined the current situation. To offset this limitation, we propose a synergy of evolutionary algorithm and reinforcement learning to investigate long-term collective performance and strategy evolution. The model adopts reinforcement learning with a tradeoff between historical and current information to make decisions when the strategies of agents evolve through repeated interactions. The results demonstrate that the strategies in populations converge to stable states, and the agents gradually form steady negotiation habits. Agents that adopt reinforcement learning perform better in payoff, fairness, and stableness than their counterparts using classic evolutionary algorithm. PMID:25048108

Evolution with reinforcement learning in negotiation.

PubMed

Zou, Yi; Zhan, Wenjie; Shao, Yuan

2014-01-01

Adaptive behavior depends less on the details of the negotiation process and makes more robust predictions in the long term as compared to in the short term. However, the extant literature on population dynamics for behavior adjustment has only examined the current situation. To offset this limitation, we propose a synergy of evolutionary algorithm and reinforcement learning to investigate long-term collective performance and strategy evolution. The model adopts reinforcement learning with a tradeoff between historical and current information to make decisions when the strategies of agents evolve through repeated interactions. The results demonstrate that the strategies in populations converge to stable states, and the agents gradually form steady negotiation habits. Agents that adopt reinforcement learning perform better in payoff, fairness, and stableness than their counterparts using classic evolutionary algorithm.

Prespeech motor learning in a neural network using reinforcement.

PubMed

Warlaumont, Anne S; Westermann, Gert; Buder, Eugene H; Oller, D Kimbrough

2013-02-01

Vocal motor development in infancy provides a crucial foundation for language development. Some significant early accomplishments include learning to control the process of phonation (the production of sound at the larynx) and learning to produce the sounds of one's language. Previous work has shown that social reinforcement shapes the kinds of vocalizations infants produce. We present a neural network model that provides an account of how vocal learning may be guided by reinforcement. The model consists of a self-organizing map that outputs to muscles of a realistic vocalization synthesizer. Vocalizations are spontaneously produced by the network. If a vocalization meets certain acoustic criteria, it is reinforced, and the weights are updated to make similar muscle activations increasingly likely to recur. We ran simulations of the model under various reinforcement criteria and tested the types of vocalizations it produced after learning in the different conditions. When reinforcement was contingent on the production of phonated (i.e. voiced) sounds, the network's post-learning productions were almost always phonated, whereas when reinforcement was not contingent on phonation, the network's post-learning productions were almost always not phonated. When reinforcement was contingent on both phonation and proximity to English vowels as opposed to Korean vowels, the model's post-learning productions were more likely to resemble the English vowels and vice versa. Copyright © 2012 Elsevier Ltd. All rights reserved.

The curse of planning: dissecting multiple reinforcement-learning systems by taxing the central executive.

PubMed

Otto, A Ross; Gershman, Samuel J; Markman, Arthur B; Daw, Nathaniel D

2013-05-01

A number of accounts of human and animal behavior posit the operation of parallel and competing valuation systems in the control of choice behavior. In these accounts, a flexible but computationally expensive model-based reinforcement-learning system has been contrasted with a less flexible but more efficient model-free reinforcement-learning system. The factors governing which system controls behavior-and under what circumstances-are still unclear. Following the hypothesis that model-based reinforcement learning requires cognitive resources, we demonstrated that having human decision makers perform a demanding secondary task engenders increased reliance on a model-free reinforcement-learning strategy. Further, we showed that, across trials, people negotiate the trade-off between the two systems dynamically as a function of concurrent executive-function demands, and people's choice latencies reflect the computational expenses of the strategy they employ. These results demonstrate that competition between multiple learning systems can be controlled on a trial-by-trial basis by modulating the availability of cognitive resources.

Multi-Objective Reinforcement Learning-based Deep Neural Networks for Cognitive Space Communications

NASA Technical Reports Server (NTRS)

Ferreria, Paulo; Paffenroth, Randy; Wyglinski, Alexander M.; Hackett, Timothy; Bilen, Sven; Reinhart, Richard; Mortensen, Dale

2017-01-01

Future communication subsystems of space exploration missions can potentially benefit from software-defined radios (SDRs) controlled by machine learning algorithms. In this paper, we propose a novel hybrid radio resource allocation management control algorithm that integrates multi-objective reinforcement learning and deep artificial neural networks. The objective is to efficiently manage communications system resources by monitoring performance functions with common dependent variables that result in conflicting goals. The uncertainty in the performance of thousands of different possible combinations of radio parameters makes the trade-off between exploration and exploitation in reinforcement learning (RL) much more challenging for future critical space-based missions. Thus, the system should spend as little time as possible on exploring actions, and whenever it explores an action, it should perform at acceptable levels most of the time. The proposed approach enables on-line learning by interactions with the environment and restricts poor resource allocation performance through virtual environment exploration. Improvements in the multiobjective performance can be achieved via transmitter parameter adaptation on a packet-basis, with poorly predicted performance promptly resulting in rejected decisions. Simulations presented in this work considered the DVB-S2 standard adaptive transmitter parameters and additional ones expected to be present in future adaptive radio systems. Performance results are provided by analysis of the proposed hybrid algorithm when operating across a satellite communication channel from Earth to GEO orbit during clear sky conditions. The proposed approach constitutes part of the core cognitive engine proof-of-concept to be delivered to the NASA Glenn Research Center SCaN Testbed located onboard the International Space Station.

Multi-Objective Reinforcement Learning-Based Deep Neural Networks for Cognitive Space Communications

NASA Technical Reports Server (NTRS)

Ferreria, Paulo Victor R.; Paffenroth, Randy; Wyglinski, Alexander M.; Hackett, Timothy M.; Bilen, Sven G.; Reinhart, Richard C.; Mortensen, Dale J.

2017-01-01

Future communication subsystems of space exploration missions can potentially benefit from software-defined radios (SDRs) controlled by machine learning algorithms. In this paper, we propose a novel hybrid radio resource allocation management control algorithm that integrates multi-objective reinforcement learning and deep artificial neural networks. The objective is to efficiently manage communications system resources by monitoring performance functions with common dependent variables that result in conflicting goals. The uncertainty in the performance of thousands of different possible combinations of radio parameters makes the trade-off between exploration and exploitation in reinforcement learning (RL) much more challenging for future critical space-based missions. Thus, the system should spend as little time as possible on exploring actions, and whenever it explores an action, it should perform at acceptable levels most of the time. The proposed approach enables on-line learning by interactions with the environment and restricts poor resource allocation performance through virtual environment exploration. Improvements in the multiobjective performance can be achieved via transmitter parameter adaptation on a packet-basis, with poorly predicted performance promptly resulting in rejected decisions. Simulations presented in this work considered the DVB-S2 standard adaptive transmitter parameters and additional ones expected to be present in future adaptive radio systems. Performance results are provided by analysis of the proposed hybrid algorithm when operating across a satellite communication channel from Earth to GEO orbit during clear sky conditions. The proposed approach constitutes part of the core cognitive engine proof-of-concept to be delivered to the NASA Glenn Research Center SCaN Testbed located onboard the International Space Station.

The drift diffusion model as the choice rule in reinforcement learning.

PubMed

Pedersen, Mads Lund; Frank, Michael J; Biele, Guido

2017-08-01

Current reinforcement-learning models often assume simplified decision processes that do not fully reflect the dynamic complexities of choice processes. Conversely, sequential-sampling models of decision making account for both choice accuracy and response time, but assume that decisions are based on static decision values. To combine these two computational models of decision making and learning, we implemented reinforcement-learning models in which the drift diffusion model describes the choice process, thereby capturing both within- and across-trial dynamics. To exemplify the utility of this approach, we quantitatively fit data from a common reinforcement-learning paradigm using hierarchical Bayesian parameter estimation, and compared model variants to determine whether they could capture the effects of stimulant medication in adult patients with attention-deficit hyperactivity disorder (ADHD). The model with the best relative fit provided a good description of the learning process, choices, and response times. A parameter recovery experiment showed that the hierarchical Bayesian modeling approach enabled accurate estimation of the model parameters. The model approach described here, using simultaneous estimation of reinforcement-learning and drift diffusion model parameters, shows promise for revealing new insights into the cognitive and neural mechanisms of learning and decision making, as well as the alteration of such processes in clinical groups.

The drift diffusion model as the choice rule in reinforcement learning

PubMed Central

Frank, Michael J.

2017-01-01

Current reinforcement-learning models often assume simplified decision processes that do not fully reflect the dynamic complexities of choice processes. Conversely, sequential-sampling models of decision making account for both choice accuracy and response time, but assume that decisions are based on static decision values. To combine these two computational models of decision making and learning, we implemented reinforcement-learning models in which the drift diffusion model describes the choice process, thereby capturing both within- and across-trial dynamics. To exemplify the utility of this approach, we quantitatively fit data from a common reinforcement-learning paradigm using hierarchical Bayesian parameter estimation, and compared model variants to determine whether they could capture the effects of stimulant medication in adult patients with attention-deficit hyper-activity disorder (ADHD). The model with the best relative fit provided a good description of the learning process, choices, and response times. A parameter recovery experiment showed that the hierarchical Bayesian modeling approach enabled accurate estimation of the model parameters. The model approach described here, using simultaneous estimation of reinforcement-learning and drift diffusion model parameters, shows promise for revealing new insights into the cognitive and neural mechanisms of learning and decision making, as well as the alteration of such processes in clinical groups. PMID:27966103

Mapping anhedonia onto reinforcement learning: a behavioural meta-analysis

PubMed Central

2013-01-01

Background Depression is characterised partly by blunted reactions to reward. However, tasks probing this deficiency have not distinguished insensitivity to reward from insensitivity to the prediction errors for reward that determine learning and are putatively reported by the phasic activity of dopamine neurons. We attempted to disentangle these factors with respect to anhedonia in the context of stress, Major Depressive Disorder (MDD), Bipolar Disorder (BPD) and a dopaminergic challenge. Methods Six behavioural datasets involving 392 experimental sessions were subjected to a model-based, Bayesian meta-analysis. Participants across all six studies performed a probabilistic reward task that used an asymmetric reinforcement schedule to assess reward learning. Healthy controls were tested under baseline conditions, stress or after receiving the dopamine D2 agonist pramipexole. In addition, participants with current or past MDD or BPD were evaluated. Reinforcement learning models isolated the contributions of variation in reward sensitivity and learning rate. Results MDD and anhedonia reduced reward sensitivity more than they affected the learning rate, while a low dose of the dopamine D2 agonist pramipexole showed the opposite pattern. Stress led to a pattern consistent with a mixed effect on reward sensitivity and learning rate. Conclusion Reward-related learning reflected at least two partially separable contributions. The first related to phasic prediction error signalling, and was preferentially modulated by a low dose of the dopamine agonist pramipexole. The second related directly to reward sensitivity, and was preferentially reduced in MDD and anhedonia. Stress altered both components. Collectively, these findings highlight the contribution of model-based reinforcement learning meta-analysis for dissecting anhedonic behavior. PMID:23782813

Working Memory and Reinforcement Schedule Jointly Determine Reinforcement Learning in Children: Potential Implications for Behavioral Parent Training

PubMed Central

Segers, Elien; Beckers, Tom; Geurts, Hilde; Claes, Laurence; Danckaerts, Marina; van der Oord, Saskia

2018-01-01

Introduction: Behavioral Parent Training (BPT) is often provided for childhood psychiatric disorders. These disorders have been shown to be associated with working memory impairments. BPT is based on operant learning principles, yet how operant principles shape behavior (through the partial reinforcement (PRF) extinction effect, i.e., greater resistance to extinction that is created when behavior is reinforced partially rather than continuously) and the potential role of working memory therein is scarcely studied in children. This study explored the PRF extinction effect and the role of working memory therein using experimental tasks in typically developing children. Methods: Ninety-seven children (age 6–10) completed a working memory task and an operant learning task, in which children acquired a response-sequence rule under either continuous or PRF (120 trials), followed by an extinction phase (80 trials). Data of 88 children were used for analysis. Results: The PRF extinction effect was confirmed: We observed slower acquisition and extinction in the PRF condition as compared to the continuous reinforcement (CRF) condition. Working memory was negatively related to acquisition but not extinction performance. Conclusion: Both reinforcement contingencies and working memory relate to acquisition performance. Potential implications for BPT are that decreasing working memory load may enhance the chance of optimally learning through reinforcement. PMID:29643822

Reinforcement learning in complementarity game and population dynamics

NASA Astrophysics Data System (ADS)

Jost, Jürgen; Li, Wei

2014-02-01

We systematically test and compare different reinforcement learning schemes in a complementarity game [J. Jost and W. Li, Physica A 345, 245 (2005), 10.1016/j.physa.2004.07.005] played between members of two populations. More precisely, we study the Roth-Erev, Bush-Mosteller, and SoftMax reinforcement learning schemes. A modified version of Roth-Erev with a power exponent of 1.5, as opposed to 1 in the standard version, performs best. We also compare these reinforcement learning strategies with evolutionary schemes. This gives insight into aspects like the issue of quick adaptation as opposed to systematic exploration or the role of learning rates.

The role of GABAB receptors in human reinforcement learning.

PubMed

Ort, Andres; Kometer, Michael; Rohde, Judith; Seifritz, Erich; Vollenweider, Franz X

2014-10-01

Behavioral evidence from human studies suggests that the γ-aminobutyric acid type B receptor (GABAB receptor) agonist baclofen modulates reinforcement learning and reduces craving in patients with addiction spectrum disorders. However, in contrast to the well established role of dopamine in reinforcement learning, the mechanisms by which the GABAB receptor influences reinforcement learning in humans remain completely unknown. To further elucidate this issue, a cross-over, double-blind, placebo-controlled study was performed in healthy human subjects (N=15) to test the effects of baclofen (20 and 50mg p.o.) on probabilistic reinforcement learning. Outcomes were the feedback-induced P2 component of the event-related potential, the feedback-related negativity, and the P300 component of the event-related potential. Baclofen produced a reduction of P2 amplitude over the course of the experiment, but did not modulate the feedback-related negativity. Furthermore, there was a trend towards increased learning after baclofen administration relative to placebo over the course of the experiment. The present results extend previous theories of reinforcement learning, which focus on the importance of mesolimbic dopamine signaling, and indicate that stimulation of cortical GABAB receptors in a fronto-parietal network leads to better attentional allocation in reinforcement learning. This observation is a first step in our understanding of how baclofen may improve reinforcement learning in healthy subjects. Further studies with bigger sample sizes are needed to corroborate this conclusion and furthermore, test this effect in patients with addiction spectrum disorder. Copyright © 2014 Elsevier B.V. and ECNP. All rights reserved.

Kernel-based least squares policy iteration for reinforcement learning.

PubMed

Xu, Xin; Hu, Dewen; Lu, Xicheng

2007-07-01

In this paper, we present a kernel-based least squares policy iteration (KLSPI) algorithm for reinforcement learning (RL) in large or continuous state spaces, which can be used to realize adaptive feedback control of uncertain dynamic systems. By using KLSPI, near-optimal control policies can be obtained without much a priori knowledge on dynamic models of control plants. In KLSPI, Mercer kernels are used in the policy evaluation of a policy iteration process, where a new kernel-based least squares temporal-difference algorithm called KLSTD-Q is proposed for efficient policy evaluation. To keep the sparsity and improve the generalization ability of KLSTD-Q solutions, a kernel sparsification procedure based on approximate linear dependency (ALD) is performed. Compared to the previous works on approximate RL methods, KLSPI makes two progresses to eliminate the main difficulties of existing results. One is the better convergence and (near) optimality guarantee by using the KLSTD-Q algorithm for policy evaluation with high precision. The other is the automatic feature selection using the ALD-based kernel sparsification. Therefore, the KLSPI algorithm provides a general RL method with generalization performance and convergence guarantee for large-scale Markov decision problems (MDPs). Experimental results on a typical RL task for a stochastic chain problem demonstrate that KLSPI can consistently achieve better learning efficiency and policy quality than the previous least squares policy iteration (LSPI) algorithm. Furthermore, the KLSPI method was also evaluated on two nonlinear feedback control problems, including a ship heading control problem and the swing up control of a double-link underactuated pendulum called acrobot. Simulation results illustrate that the proposed method can optimize controller performance using little a priori information of uncertain dynamic systems. It is also demonstrated that KLSPI can be applied to online learning control by incorporating

Separation of Time-Based and Trial-Based Accounts of the Partial Reinforcement Extinction Effect

PubMed Central

Bouton, Mark E.; Woods, Amanda M.; Todd, Travis P.

2013-01-01

Two appetitive conditioning experiments with rats examined time-based and trial-based accounts of the partial reinforcement extinction effect (PREE). In the PREE, the loss of responding that occurs in extinction is slower when the conditioned stimulus (CS) has been paired with a reinforcer on some of its presentations (partially reinforced) instead of every presentation (continuously reinforced). According to a time-based or “time-accumulation” view (e.g., Gallistel & Gibbon, 2000), the PREE occurs because the organism has learned in partial reinforcement to expect the reinforcer after a larger amount of time has accumulated in the CS over trials. In contrast, according to a trial-based view (e.g., Capaldi, 1967), the PREE occurs because the organism has learned in partial reinforcement to expect the reinforcer after a larger number of CS presentations. Experiment 1 used a procedure that equated partially- and continuously-reinforced groups on their expected times to reinforcement during conditioning. A PREE was still observed. Experiment 2 then used an extinction procedure that allowed time in the CS and the number of trials to accumulate differentially through extinction. The PREE was still evident when responding was examined as a function of expected time units to the reinforcer, but was eliminated when responding was examined as a function of expected trial units to the reinforcer. There was no evidence that the animal responded according to the ratio of time accumulated during the CS in extinction over the time in the CS expected before the reinforcer. The results thus favor a trial-based account over a time-based account of extinction and the PREE. PMID:23962669

Punishment Insensitivity and Impaired Reinforcement Learning in Preschoolers

ERIC Educational Resources Information Center

Briggs-Gowan, Margaret J.; Nichols, Sara R.; Voss, Joel; Zobel, Elvira; Carter, Alice S.; McCarthy, Kimberly J.; Pine, Daniel S.; Blair, James; Wakschlag, Lauren S.

2014-01-01

Background: Youth and adults with psychopathic traits display disrupted reinforcement learning. Advances in measurement now enable examination of this association in preschoolers. The current study examines relations between reinforcement learning in preschoolers and parent ratings of reduced responsiveness to socialization, conceptualized as a…

Reinforcement active learning in the vibrissae system: optimal object localization.

PubMed

Gordon, Goren; Dorfman, Nimrod; Ahissar, Ehud

2013-01-01

Rats move their whiskers to acquire information about their environment. It has been observed that they palpate novel objects and objects they are required to localize in space. We analyze whisker-based object localization using two complementary paradigms, namely, active learning and intrinsic-reward reinforcement learning. Active learning algorithms select the next training samples according to the hypothesized solution in order to better discriminate between correct and incorrect labels. Intrinsic-reward reinforcement learning uses prediction errors as the reward to an actor-critic design, such that behavior converges to the one that optimizes the learning process. We show that in the context of object localization, the two paradigms result in palpation whisking as their respective optimal solution. These results suggest that rats may employ principles of active learning and/or intrinsic reward in tactile exploration and can guide future research to seek the underlying neuronal mechanisms that implement them. Furthermore, these paradigms are easily transferable to biomimetic whisker-based artificial sensors and can improve the active exploration of their environment. Copyright © 2012 Elsevier Ltd. All rights reserved.

Instructional control of reinforcement learning: A behavioral and neurocomputational investigation

PubMed Central

Doll, Bradley B.; Jacobs, W. Jake; Sanfey, Alan G.; Frank, Michael J.

2011-01-01

Humans learn how to behave directly through environmental experience and indirectly through rules and instructions. Behavior analytic research has shown that instructions can control behavior, even when such behavior leads to sub-optimal outcomes (Hayes, S. (Ed.). 1989. Rule-governed behavior: cognition, contingencies, and instructional control. Plenum Press.). Here we examine the control of behavior through instructions in a reinforcement learning task known to depend on striatal dopaminergic function. Participants selected between probabilistically reinforced stimuli, and were (incorrectly) told that a specific stimulus had the highest (or lowest) reinforcement probability. Despite experience to the contrary, instructions drove choice behavior. We present neural network simulations that capture the interactions between instruction-driven and reinforcement-driven behavior via two potential neural circuits: one in which the striatum is inaccurately trained by instruction representations coming from prefrontal cortex/hippocampus (PFC/HC), and another in which the striatum learns the environmentally based reinforcement contingencies, but is “overridden” at decision output. Both models capture the core behavioral phenomena but, because they differ fundamentally on what is learned, make distinct predictions for subsequent behavioral and neuroimaging experiments. Finally, we attempt to distinguish between the proposed computational mechanisms governing instructed behavior by fitting a series of abstract “Q-learning” and Bayesian models to subject data. The best-fitting model supports one of the neural models, suggesting the existence of a “confirmation bias” in which the PFC/HC system trains the reinforcement system by amplifying outcomes that are consistent with instructions while diminishing inconsistent outcomes. PMID:19595993

Neural Basis of Reinforcement Learning and Decision Making

PubMed Central

Lee, Daeyeol; Seo, Hyojung; Jung, Min Whan

2012-01-01

Reinforcement learning is an adaptive process in which an animal utilizes its previous experience to improve the outcomes of future choices. Computational theories of reinforcement learning play a central role in the newly emerging areas of neuroeconomics and decision neuroscience. In this framework, actions are chosen according to their value functions, which describe how much future reward is expected from each action. Value functions can be adjusted not only through reward and penalty, but also by the animal’s knowledge of its current environment. Studies have revealed that a large proportion of the brain is involved in representing and updating value functions and using them to choose an action. However, how the nature of a behavioral task affects the neural mechanisms of reinforcement learning remains incompletely understood. Future studies should uncover the principles by which different computational elements of reinforcement learning are dynamically coordinated across the entire brain. PMID:22462543

Reinforcement learning improves behaviour from evaluative feedback

NASA Astrophysics Data System (ADS)

Littman, Michael L.

2015-05-01

Reinforcement learning is a branch of machine learning concerned with using experience gained through interacting with the world and evaluative feedback to improve a system's ability to make behavioural decisions. It has been called the artificial intelligence problem in a microcosm because learning algorithms must act autonomously to perform well and achieve their goals. Partly driven by the increasing availability of rich data, recent years have seen exciting advances in the theory and practice of reinforcement learning, including developments in fundamental technical areas such as generalization, planning, exploration and empirical methodology, leading to increasing applicability to real-life problems.

Reinforcement learning improves behaviour from evaluative feedback.

PubMed

Littman, Michael L

2015-05-28

Reinforcement learning is a branch of machine learning concerned with using experience gained through interacting with the world and evaluative feedback to improve a system's ability to make behavioural decisions. It has been called the artificial intelligence problem in a microcosm because learning algorithms must act autonomously to perform well and achieve their goals. Partly driven by the increasing availability of rich data, recent years have seen exciting advances in the theory and practice of reinforcement learning, including developments in fundamental technical areas such as generalization, planning, exploration and empirical methodology, leading to increasing applicability to real-life problems.

Reinforcement Learning and Dopamine in Schizophrenia: Dimensions of Symptoms or Specific Features of a Disease Group?

PubMed Central

Deserno, Lorenz; Boehme, Rebecca; Heinz, Andreas; Schlagenhauf, Florian

2013-01-01

Abnormalities in reinforcement learning are a key finding in schizophrenia and have been proposed to be linked to elevated levels of dopamine neurotransmission. Behavioral deficits in reinforcement learning and their neural correlates may contribute to the formation of clinical characteristics of schizophrenia. The ability to form predictions about future outcomes is fundamental for environmental interactions and depends on neuronal teaching signals, like reward prediction errors. While aberrant prediction errors, that encode non-salient events as surprising, have been proposed to contribute to the formation of positive symptoms, a failure to build neural representations of decision values may result in negative symptoms. Here, we review behavioral and neuroimaging research in schizophrenia and focus on studies that implemented reinforcement learning models. In addition, we discuss studies that combined reinforcement learning with measures of dopamine. Thereby, we suggest how reinforcement learning abnormalities in schizophrenia may contribute to the formation of psychotic symptoms and may interact with cognitive deficits. These ideas point toward an interplay of more rigid versus flexible control over reinforcement learning. Pronounced deficits in the flexible or model-based domain may allow for a detailed characterization of well-established cognitive deficits in schizophrenia patients based on computational models of learning. Finally, we propose a framework based on the potentially crucial contribution of dopamine to dysfunctional reinforcement learning on the level of neural networks. Future research may strongly benefit from computational modeling but also requires further methodological improvement for clinical group studies. These research tools may help to improve our understanding of disease-specific mechanisms and may help to identify clinically relevant subgroups of the heterogeneous entity schizophrenia. PMID:24391603

Cognitive control predicts use of model-based reinforcement learning.

PubMed

Otto, A Ross; Skatova, Anya; Madlon-Kay, Seth; Daw, Nathaniel D

2015-02-01

Accounts of decision-making and its neural substrates have long posited the operation of separate, competing valuation systems in the control of choice behavior. Recent theoretical and experimental work suggest that this classic distinction between behaviorally and neurally dissociable systems for habitual and goal-directed (or more generally, automatic and controlled) choice may arise from two computational strategies for reinforcement learning (RL), called model-free and model-based RL, but the cognitive or computational processes by which one system may dominate over the other in the control of behavior is a matter of ongoing investigation. To elucidate this question, we leverage the theoretical framework of cognitive control, demonstrating that individual differences in utilization of goal-related contextual information--in the service of overcoming habitual, stimulus-driven responses--in established cognitive control paradigms predict model-based behavior in a separate, sequential choice task. The behavioral correspondence between cognitive control and model-based RL compellingly suggests that a common set of processes may underpin the two behaviors. In particular, computational mechanisms originally proposed to underlie controlled behavior may be applicable to understanding the interactions between model-based and model-free choice behavior.

Cognitive Control Predicts Use of Model-Based Reinforcement-Learning

PubMed Central

Otto, A. Ross; Skatova, Anya; Madlon-Kay, Seth; Daw, Nathaniel D.

2015-01-01

Accounts of decision-making and its neural substrates have long posited the operation of separate, competing valuation systems in the control of choice behavior. Recent theoretical and experimental work suggest that this classic distinction between behaviorally and neurally dissociable systems for habitual and goal-directed (or more generally, automatic and controlled) choice may arise from two computational strategies for reinforcement learning (RL), called model-free and model-based RL, but the cognitive or computational processes by which one system may dominate over the other in the control of behavior is a matter of ongoing investigation. To elucidate this question, we leverage the theoretical framework of cognitive control, demonstrating that individual differences in utilization of goal-related contextual information—in the service of overcoming habitual, stimulus-driven responses—in established cognitive control paradigms predict model-based behavior in a separate, sequential choice task. The behavioral correspondence between cognitive control and model-based RL compellingly suggests that a common set of processes may underpin the two behaviors. In particular, computational mechanisms originally proposed to underlie controlled behavior may be applicable to understanding the interactions between model-based and model-free choice behavior. PMID:25170791

Reinforcement learning of periodical gaits in locomotion robots

NASA Astrophysics Data System (ADS)

Svinin, Mikhail; Yamada, Kazuyaki; Ushio, S.; Ueda, Kanji

1999-08-01

Emergence of stable gaits in locomotion robots is studied in this paper. A classifier system, implementing an instance- based reinforcement learning scheme, is used for sensory- motor control of an eight-legged mobile robot. Important feature of the classifier system is its ability to work with the continuous sensor space. The robot does not have a prior knowledge of the environment, its own internal model, and the goal coordinates. It is only assumed that the robot can acquire stable gaits by learning how to reach a light source. During the learning process the control system, is self-organized by reinforcement signals. Reaching the light source defines a global reward. Forward motion gets a local reward, while stepping back and falling down get a local punishment. Feasibility of the proposed self-organized system is tested under simulation and experiment. The control actions are specified at the leg level. It is shown that, as learning progresses, the number of the action rules in the classifier systems is stabilized to a certain level, corresponding to the acquired gait patterns.

Joint Extraction of Entities and Relations Using Reinforcement Learning and Deep Learning.

PubMed

Feng, Yuntian; Zhang, Hongjun; Hao, Wenning; Chen, Gang

2017-01-01

We use both reinforcement learning and deep learning to simultaneously extract entities and relations from unstructured texts. For reinforcement learning, we model the task as a two-step decision process. Deep learning is used to automatically capture the most important information from unstructured texts, which represent the state in the decision process. By designing the reward function per step, our proposed method can pass the information of entity extraction to relation extraction and obtain feedback in order to extract entities and relations simultaneously. Firstly, we use bidirectional LSTM to model the context information, which realizes preliminary entity extraction. On the basis of the extraction results, attention based method can represent the sentences that include target entity pair to generate the initial state in the decision process. Then we use Tree-LSTM to represent relation mentions to generate the transition state in the decision process. Finally, we employ Q -Learning algorithm to get control policy π in the two-step decision process. Experiments on ACE2005 demonstrate that our method attains better performance than the state-of-the-art method and gets a 2.4% increase in recall-score.

Joint Extraction of Entities and Relations Using Reinforcement Learning and Deep Learning

PubMed Central

Zhang, Hongjun; Chen, Gang

2017-01-01

We use both reinforcement learning and deep learning to simultaneously extract entities and relations from unstructured texts. For reinforcement learning, we model the task as a two-step decision process. Deep learning is used to automatically capture the most important information from unstructured texts, which represent the state in the decision process. By designing the reward function per step, our proposed method can pass the information of entity extraction to relation extraction and obtain feedback in order to extract entities and relations simultaneously. Firstly, we use bidirectional LSTM to model the context information, which realizes preliminary entity extraction. On the basis of the extraction results, attention based method can represent the sentences that include target entity pair to generate the initial state in the decision process. Then we use Tree-LSTM to represent relation mentions to generate the transition state in the decision process. Finally, we employ Q-Learning algorithm to get control policy π in the two-step decision process. Experiments on ACE2005 demonstrate that our method attains better performance than the state-of-the-art method and gets a 2.4% increase in recall-score. PMID:28894463

Hierarchically organized behavior and its neural foundations: A reinforcement-learning perspective

PubMed Central

Botvinick, Matthew M.; Niv, Yael; Barto, Andrew C.

2009-01-01

Research on human and animal behavior has long emphasized its hierarchical structure — the divisibility of ongoing behavior into discrete tasks, which are comprised of subtask sequences, which in turn are built of simple actions. The hierarchical structure of behavior has also been of enduring interest within neuroscience, where it has been widely considered to reflect prefrontal cortical functions. In this paper, we reexamine behavioral hierarchy and its neural substrates from the point of view of recent developments in computational reinforcement learning. Specifically, we consider a set of approaches known collectively as hierarchical reinforcement learning, which extend the reinforcement learning paradigm by allowing the learning agent to aggregate actions into reusable subroutines or skills. A close look at the components of hierarchical reinforcement learning suggests how they might map onto neural structures, in particular regions within the dorsolateral and orbital prefrontal cortex. It also suggests specific ways in which hierarchical reinforcement learning might provide a complement to existing psychological models of hierarchically structured behavior. A particularly important question that hierarchical reinforcement learning brings to the fore is that of how learning identifies new action routines that are likely to provide useful building blocks in solving a wide range of future problems. Here and at many other points, hierarchical reinforcement learning offers an appealing framework for investigating the computational and neural underpinnings of hierarchically structured behavior. PMID:18926527

Utilising reinforcement learning to develop strategies for driving auditory neural implants.

PubMed

Lee, Geoffrey W; Zambetta, Fabio; Li, Xiaodong; Paolini, Antonio G

2016-08-01

In this paper we propose a novel application of reinforcement learning to the area of auditory neural stimulation. We aim to develop a simulation environment which is based off real neurological responses to auditory and electrical stimulation in the cochlear nucleus (CN) and inferior colliculus (IC) of an animal model. Using this simulator we implement closed loop reinforcement learning algorithms to determine which methods are most effective at learning effective acoustic neural stimulation strategies. By recording a comprehensive set of acoustic frequency presentations and neural responses from a set of animals we created a large database of neural responses to acoustic stimulation. Extensive electrical stimulation in the CN and the recording of neural responses in the IC provides a mapping of how the auditory system responds to electrical stimuli. The combined dataset is used as the foundation for the simulator, which is used to implement and test learning algorithms. Reinforcement learning, utilising a modified n-Armed Bandit solution, is implemented to demonstrate the model's function. We show the ability to effectively learn stimulation patterns which mimic the cochlea's ability to covert acoustic frequencies to neural activity. Time taken to learn effective replication using neural stimulation takes less than 20 min under continuous testing. These results show the utility of reinforcement learning in the field of neural stimulation. These results can be coupled with existing sound processing technologies to develop new auditory prosthetics that are adaptable to the recipients current auditory pathway. The same process can theoretically be abstracted to other sensory and motor systems to develop similar electrical replication of neural signals.

Cardiac Concomitants of Feedback and Prediction Error Processing in Reinforcement Learning.

PubMed

Kastner, Lucas; Kube, Jana; Villringer, Arno; Neumann, Jane

2017-01-01

Successful learning hinges on the evaluation of positive and negative feedback. We assessed differential learning from reward and punishment in a monetary reinforcement learning paradigm, together with cardiac concomitants of positive and negative feedback processing. On the behavioral level, learning from reward resulted in more advantageous behavior than learning from punishment, suggesting a differential impact of reward and punishment on successful feedback-based learning. On the autonomic level, learning and feedback processing were closely mirrored by phasic cardiac responses on a trial-by-trial basis: (1) Negative feedback was accompanied by faster and prolonged heart rate deceleration compared to positive feedback. (2) Cardiac responses shifted from feedback presentation at the beginning of learning to stimulus presentation later on. (3) Most importantly, the strength of phasic cardiac responses to the presentation of feedback correlated with the strength of prediction error signals that alert the learner to the necessity for behavioral adaptation. Considering participants' weight status and gender revealed obesity-related deficits in learning to avoid negative consequences and less consistent behavioral adaptation in women compared to men. In sum, our results provide strong new evidence for the notion that during learning phasic cardiac responses reflect an internal value and feedback monitoring system that is sensitive to the violation of performance-based expectations. Moreover, inter-individual differences in weight status and gender may affect both behavioral and autonomic responses in reinforcement-based learning.

Variability in Dopamine Genes Dissociates Model-Based and Model-Free Reinforcement Learning.

PubMed

Doll, Bradley B; Bath, Kevin G; Daw, Nathaniel D; Frank, Michael J

2016-01-27

Considerable evidence suggests that multiple learning systems can drive behavior. Choice can proceed reflexively from previous actions and their associated outcomes, as captured by "model-free" learning algorithms, or flexibly from prospective consideration of outcomes that might occur, as captured by "model-based" learning algorithms. However, differential contributions of dopamine to these systems are poorly understood. Dopamine is widely thought to support model-free learning by modulating plasticity in striatum. Model-based learning may also be affected by these striatal effects, or by other dopaminergic effects elsewhere, notably on prefrontal working memory function. Indeed, prominent demonstrations linking striatal dopamine to putatively model-free learning did not rule out model-based effects, whereas other studies have reported dopaminergic modulation of verifiably model-based learning, but without distinguishing a prefrontal versus striatal locus. To clarify the relationships between dopamine, neural systems, and learning strategies, we combine a genetic association approach in humans with two well-studied reinforcement learning tasks: one isolating model-based from model-free behavior and the other sensitive to key aspects of striatal plasticity. Prefrontal function was indexed by a polymorphism in the COMT gene, differences of which reflect dopamine levels in the prefrontal cortex. This polymorphism has been associated with differences in prefrontal activity and working memory. Striatal function was indexed by a gene coding for DARPP-32, which is densely expressed in the striatum where it is necessary for synaptic plasticity. We found evidence for our hypothesis that variations in prefrontal dopamine relate to model-based learning, whereas variations in striatal dopamine function relate to model-free learning. Decisions can stem reflexively from their previously associated outcomes or flexibly from deliberative consideration of potential choice outcomes

A computational psychiatry approach identifies how alpha-2A noradrenergic agonist Guanfacine affects feature-based reinforcement learning in the macaque

PubMed Central

Hassani, S. A.; Oemisch, M.; Balcarras, M.; Westendorff, S.; Ardid, S.; van der Meer, M. A.; Tiesinga, P.; Womelsdorf, T.

2017-01-01

Noradrenaline is believed to support cognitive flexibility through the alpha 2A noradrenergic receptor (a2A-NAR) acting in prefrontal cortex. Enhanced flexibility has been inferred from improved working memory with the a2A-NA agonist Guanfacine. But it has been unclear whether Guanfacine improves specific attention and learning mechanisms beyond working memory, and whether the drug effects can be formalized computationally to allow single subject predictions. We tested and confirmed these suggestions in a case study with a healthy nonhuman primate performing a feature-based reversal learning task evaluating performance using Bayesian and Reinforcement learning models. In an initial dose-testing phase we found a Guanfacine dose that increased performance accuracy, decreased distractibility and improved learning. In a second experimental phase using only that dose we examined the faster feature-based reversal learning with Guanfacine with single-subject computational modeling. Parameter estimation suggested that improved learning is not accounted for by varying a single reinforcement learning mechanism, but by changing the set of parameter values to higher learning rates and stronger suppression of non-chosen over chosen feature information. These findings provide an important starting point for developing nonhuman primate models to discern the synaptic mechanisms of attention and learning functions within the context of a computational neuropsychiatry framework. PMID:28091572

Changes in corticostriatal connectivity during reinforcement learning in humans.

PubMed

Horga, Guillermo; Maia, Tiago V; Marsh, Rachel; Hao, Xuejun; Xu, Dongrong; Duan, Yunsuo; Tau, Gregory Z; Graniello, Barbara; Wang, Zhishun; Kangarlu, Alayar; Martinez, Diana; Packard, Mark G; Peterson, Bradley S

2015-02-01

Many computational models assume that reinforcement learning relies on changes in synaptic efficacy between cortical regions representing stimuli and striatal regions involved in response selection, but this assumption has thus far lacked empirical support in humans. We recorded hemodynamic signals with fMRI while participants navigated a virtual maze to find hidden rewards. We fitted a reinforcement-learning algorithm to participants' choice behavior and evaluated the neural activity and the changes in functional connectivity related to trial-by-trial learning variables. Activity in the posterior putamen during choice periods increased progressively during learning. Furthermore, the functional connections between the sensorimotor cortex and the posterior putamen strengthened progressively as participants learned the task. These changes in corticostriatal connectivity differentiated participants who learned the task from those who did not. These findings provide a direct link between changes in corticostriatal connectivity and learning, thereby supporting a central assumption common to several computational models of reinforcement learning. © 2014 Wiley Periodicals, Inc.

Regulating recognition decisions through incremental reinforcement learning.

PubMed

Han, Sanghoon; Dobbins, Ian G

2009-06-01

Does incremental reinforcement learning influence recognition memory judgments? We examined this question by subtly altering the relative validity or availability of feedback in order to differentially reinforce old or new recognition judgments. Experiment 1 probabilistically and incorrectly indicated that either misses or false alarms were correct in the context of feedback that was otherwise accurate. Experiment 2 selectively withheld feedback for either misses or false alarms in the context of feedback that was otherwise present. Both manipulations caused prominent shifts of recognition memory decision criteria that remained for considerable periods even after feedback had been altogether removed. Overall, these data demonstrate that incremental reinforcement-learning mechanisms influence the degree of caution subjects exercise when evaluating explicit memories.

Gaze data reveal distinct choice processes underlying model-based and model-free reinforcement learning

PubMed Central

Konovalov, Arkady; Krajbich, Ian

2016-01-01

Organisms appear to learn and make decisions using different strategies known as model-free and model-based learning; the former is mere reinforcement of previously rewarded actions and the latter is a forward-looking strategy that involves evaluation of action-state transition probabilities. Prior work has used neural data to argue that both model-based and model-free learners implement a value comparison process at trial onset, but model-based learners assign more weight to forward-looking computations. Here using eye-tracking, we report evidence for a different interpretation of prior results: model-based subjects make their choices prior to trial onset. In contrast, model-free subjects tend to ignore model-based aspects of the task and instead seem to treat the decision problem as a simple comparison process between two differentially valued items, consistent with previous work on sequential-sampling models of decision making. These findings illustrate a problem with assuming that experimental subjects make their decisions at the same prescribed time. PMID:27511383

Reinforcement learning or active inference?

PubMed

Friston, Karl J; Daunizeau, Jean; Kiebel, Stefan J

2009-07-29

This paper questions the need for reinforcement learning or control theory when optimising behaviour. We show that it is fairly simple to teach an agent complicated and adaptive behaviours using a free-energy formulation of perception. In this formulation, agents adjust their internal states and sampling of the environment to minimize their free-energy. Such agents learn causal structure in the environment and sample it in an adaptive and self-supervised fashion. This results in behavioural policies that reproduce those optimised by reinforcement learning and dynamic programming. Critically, we do not need to invoke the notion of reward, value or utility. We illustrate these points by solving a benchmark problem in dynamic programming; namely the mountain-car problem, using active perception or inference under the free-energy principle. The ensuing proof-of-concept may be important because the free-energy formulation furnishes a unified account of both action and perception and may speak to a reappraisal of the role of dopamine in the brain.

A Flexible Mechanism of Rule Selection Enables Rapid Feature-Based Reinforcement Learning

PubMed Central

Balcarras, Matthew; Womelsdorf, Thilo

2016-01-01

Learning in a new environment is influenced by prior learning and experience. Correctly applying a rule that maps a context to stimuli, actions, and outcomes enables faster learning and better outcomes compared to relying on strategies for learning that are ignorant of task structure. However, it is often difficult to know when and how to apply learned rules in new contexts. In our study we explored how subjects employ different strategies for learning the relationship between stimulus features and positive outcomes in a probabilistic task context. We test the hypothesis that task naive subjects will show enhanced learning of feature specific reward associations by switching to the use of an abstract rule that associates stimuli by feature type and restricts selections to that dimension. To test this hypothesis we designed a decision making task where subjects receive probabilistic feedback following choices between pairs of stimuli. In the task, trials are grouped in two contexts by blocks, where in one type of block there is no unique relationship between a specific feature dimension (stimulus shape or color) and positive outcomes, and following an un-cued transition, alternating blocks have outcomes that are linked to either stimulus shape or color. Two-thirds of subjects (n = 22/32) exhibited behavior that was best fit by a hierarchical feature-rule model. Supporting the prediction of the model mechanism these subjects showed significantly enhanced performance in feature-reward blocks, and rapidly switched their choice strategy to using abstract feature rules when reward contingencies changed. Choice behavior of other subjects (n = 10/32) was fit by a range of alternative reinforcement learning models representing strategies that do not benefit from applying previously learned rules. In summary, these results show that untrained subjects are capable of flexibly shifting between behavioral rules by leveraging simple model-free reinforcement learning and context

Variability in Dopamine Genes Dissociates Model-Based and Model-Free Reinforcement Learning

PubMed Central

Bath, Kevin G.; Daw, Nathaniel D.; Frank, Michael J.

2016-01-01

Considerable evidence suggests that multiple learning systems can drive behavior. Choice can proceed reflexively from previous actions and their associated outcomes, as captured by “model-free” learning algorithms, or flexibly from prospective consideration of outcomes that might occur, as captured by “model-based” learning algorithms. However, differential contributions of dopamine to these systems are poorly understood. Dopamine is widely thought to support model-free learning by modulating plasticity in striatum. Model-based learning may also be affected by these striatal effects, or by other dopaminergic effects elsewhere, notably on prefrontal working memory function. Indeed, prominent demonstrations linking striatal dopamine to putatively model-free learning did not rule out model-based effects, whereas other studies have reported dopaminergic modulation of verifiably model-based learning, but without distinguishing a prefrontal versus striatal locus. To clarify the relationships between dopamine, neural systems, and learning strategies, we combine a genetic association approach in humans with two well-studied reinforcement learning tasks: one isolating model-based from model-free behavior and the other sensitive to key aspects of striatal plasticity. Prefrontal function was indexed by a polymorphism in the COMT gene, differences of which reflect dopamine levels in the prefrontal cortex. This polymorphism has been associated with differences in prefrontal activity and working memory. Striatal function was indexed by a gene coding for DARPP-32, which is densely expressed in the striatum where it is necessary for synaptic plasticity. We found evidence for our hypothesis that variations in prefrontal dopamine relate to model-based learning, whereas variations in striatal dopamine function relate to model-free learning. SIGNIFICANCE STATEMENT Decisions can stem reflexively from their previously associated outcomes or flexibly from deliberative

Neural correlates of reinforcement learning and social preferences in competitive bidding.

PubMed

van den Bos, Wouter; Talwar, Arjun; McClure, Samuel M

2013-01-30

In competitive social environments, people often deviate from what rational choice theory prescribes, resulting in losses or suboptimal monetary gains. We investigate how competition affects learning and decision-making in a common value auction task. During the experiment, groups of five human participants were simultaneously scanned using MRI while playing the auction task. We first demonstrate that bidding is well characterized by reinforcement learning with biased reward representations dependent on social preferences. Indicative of reinforcement learning, we found that estimated trial-by-trial prediction errors correlated with activity in the striatum and ventromedial prefrontal cortex. Additionally, we found that individual differences in social preferences were related to activity in the temporal-parietal junction and anterior insula. Connectivity analyses suggest that monetary and social value signals are integrated in the ventromedial prefrontal cortex and striatum. Based on these results, we argue for a novel mechanistic account for the integration of reinforcement history and social preferences in competitive decision-making.

Reinforcement Learning in Multidimensional Environments Relies on Attention Mechanisms

PubMed Central

Daniel, Reka; Geana, Andra; Gershman, Samuel J.; Leong, Yuan Chang; Radulescu, Angela; Wilson, Robert C.

2015-01-01

In recent years, ideas from the computational field of reinforcement learning have revolutionized the study of learning in the brain, famously providing new, precise theories of how dopamine affects learning in the basal ganglia. However, reinforcement learning algorithms are notorious for not scaling well to multidimensional environments, as is required for real-world learning. We hypothesized that the brain naturally reduces the dimensionality of real-world problems to only those dimensions that are relevant to predicting reward, and conducted an experiment to assess by what algorithms and with what neural mechanisms this “representation learning” process is realized in humans. Our results suggest that a bilateral attentional control network comprising the intraparietal sulcus, precuneus, and dorsolateral prefrontal cortex is involved in selecting what dimensions are relevant to the task at hand, effectively updating the task representation through trial and error. In this way, cortical attention mechanisms interact with learning in the basal ganglia to solve the “curse of dimensionality” in reinforcement learning. PMID:26019331

Social Cognition as Reinforcement Learning: Feedback Modulates Emotion Inference.

PubMed

Zaki, Jamil; Kallman, Seth; Wimmer, G Elliott; Ochsner, Kevin; Shohamy, Daphna

2016-09-01

Neuroscientific studies of social cognition typically employ paradigms in which perceivers draw single-shot inferences about the internal states of strangers. Real-world social inference features much different parameters: People often encounter and learn about particular social targets (e.g., friends) over time and receive feedback about whether their inferences are correct or incorrect. Here, we examined this process and, more broadly, the intersection between social cognition and reinforcement learning. Perceivers were scanned using fMRI while repeatedly encountering three social targets who produced conflicting visual and verbal emotional cues. Perceivers guessed how targets felt and received feedback about whether they had guessed correctly. Visual cues reliably predicted one target's emotion, verbal cues predicted a second target's emotion, and neither reliably predicted the third target's emotion. Perceivers successfully used this information to update their judgments over time. Furthermore, trial-by-trial learning signals-estimated using two reinforcement learning models-tracked activity in ventral striatum and ventromedial pFC, structures associated with reinforcement learning, and regions associated with updating social impressions, including TPJ. These data suggest that learning about others' emotions, like other forms of feedback learning, relies on domain-general reinforcement mechanisms as well as domain-specific social information processing.

Cardiac Concomitants of Feedback and Prediction Error Processing in Reinforcement Learning

PubMed Central

Kastner, Lucas; Kube, Jana; Villringer, Arno; Neumann, Jane

2017-01-01

Successful learning hinges on the evaluation of positive and negative feedback. We assessed differential learning from reward and punishment in a monetary reinforcement learning paradigm, together with cardiac concomitants of positive and negative feedback processing. On the behavioral level, learning from reward resulted in more advantageous behavior than learning from punishment, suggesting a differential impact of reward and punishment on successful feedback-based learning. On the autonomic level, learning and feedback processing were closely mirrored by phasic cardiac responses on a trial-by-trial basis: (1) Negative feedback was accompanied by faster and prolonged heart rate deceleration compared to positive feedback. (2) Cardiac responses shifted from feedback presentation at the beginning of learning to stimulus presentation later on. (3) Most importantly, the strength of phasic cardiac responses to the presentation of feedback correlated with the strength of prediction error signals that alert the learner to the necessity for behavioral adaptation. Considering participants' weight status and gender revealed obesity-related deficits in learning to avoid negative consequences and less consistent behavioral adaptation in women compared to men. In sum, our results provide strong new evidence for the notion that during learning phasic cardiac responses reflect an internal value and feedback monitoring system that is sensitive to the violation of performance-based expectations. Moreover, inter-individual differences in weight status and gender may affect both behavioral and autonomic responses in reinforcement-based learning. PMID:29163004

Reinforcement learning for a biped robot based on a CPG-actor-critic method.

PubMed

Nakamura, Yutaka; Mori, Takeshi; Sato, Masa-aki; Ishii, Shin

2007-08-01

Animals' rhythmic movements, such as locomotion, are considered to be controlled by neural circuits called central pattern generators (CPGs), which generate oscillatory signals. Motivated by this biological mechanism, studies have been conducted on the rhythmic movements controlled by CPG. As an autonomous learning framework for a CPG controller, we propose in this article a reinforcement learning method we call the "CPG-actor-critic" method. This method introduces a new architecture to the actor, and its training is roughly based on a stochastic policy gradient algorithm presented recently. We apply this method to an automatic acquisition problem of control for a biped robot. Computer simulations show that training of the CPG can be successfully performed by our method, thus allowing the biped robot to not only walk stably but also adapt to environmental changes.

Reinforcement learning in multidimensional environments relies on attention mechanisms.

PubMed

Niv, Yael; Daniel, Reka; Geana, Andra; Gershman, Samuel J; Leong, Yuan Chang; Radulescu, Angela; Wilson, Robert C

2015-05-27

In recent years, ideas from the computational field of reinforcement learning have revolutionized the study of learning in the brain, famously providing new, precise theories of how dopamine affects learning in the basal ganglia. However, reinforcement learning algorithms are notorious for not scaling well to multidimensional environments, as is required for real-world learning. We hypothesized that the brain naturally reduces the dimensionality of real-world problems to only those dimensions that are relevant to predicting reward, and conducted an experiment to assess by what algorithms and with what neural mechanisms this "representation learning" process is realized in humans. Our results suggest that a bilateral attentional control network comprising the intraparietal sulcus, precuneus, and dorsolateral prefrontal cortex is involved in selecting what dimensions are relevant to the task at hand, effectively updating the task representation through trial and error. In this way, cortical attention mechanisms interact with learning in the basal ganglia to solve the "curse of dimensionality" in reinforcement learning. Copyright © 2015 the authors 0270-6474/15/358145-13$15.00/0.

Novelty and Inductive Generalization in Human Reinforcement Learning

PubMed Central

Gershman, Samuel J.; Niv, Yael

2015-01-01

In reinforcement learning, a decision maker searching for the most rewarding option is often faced with the question: what is the value of an option that has never been tried before? One way to frame this question is as an inductive problem: how can I generalize my previous experience with one set of options to a novel option? We show how hierarchical Bayesian inference can be used to solve this problem, and describe an equivalence between the Bayesian model and temporal difference learning algorithms that have been proposed as models of reinforcement learning in humans and animals. According to our view, the search for the best option is guided by abstract knowledge about the relationships between different options in an environment, resulting in greater search efficiency compared to traditional reinforcement learning algorithms previously applied to human cognition. In two behavioral experiments, we test several predictions of our model, providing evidence that humans learn and exploit structured inductive knowledge to make predictions about novel options. In light of this model, we suggest a new interpretation of dopaminergic responses to novelty. PMID:25808176

A graph-based evolutionary algorithm: Genetic Network Programming (GNP) and its extension using reinforcement learning.

PubMed

Mabu, Shingo; Hirasawa, Kotaro; Hu, Jinglu

2007-01-01

This paper proposes a graph-based evolutionary algorithm called Genetic Network Programming (GNP). Our goal is to develop GNP, which can deal with dynamic environments efficiently and effectively, based on the distinguished expression ability of the graph (network) structure. The characteristics of GNP are as follows. 1) GNP programs are composed of a number of nodes which execute simple judgment/processing, and these nodes are connected by directed links to each other. 2) The graph structure enables GNP to re-use nodes, thus the structure can be very compact. 3) The node transition of GNP is executed according to its node connections without any terminal nodes, thus the past history of the node transition affects the current node to be used and this characteristic works as an implicit memory function. These structural characteristics are useful for dealing with dynamic environments. Furthermore, we propose an extended algorithm, "GNP with Reinforcement Learning (GNPRL)" which combines evolution and reinforcement learning in order to create effective graph structures and obtain better results in dynamic environments. In this paper, we applied GNP to the problem of determining agents' behavior to evaluate its effectiveness. Tileworld was used as the simulation environment. The results show some advantages for GNP over conventional methods.

Punishment insensitivity and impaired reinforcement learning in preschoolers.

PubMed

Briggs-Gowan, Margaret J; Nichols, Sara R; Voss, Joel; Zobel, Elvira; Carter, Alice S; McCarthy, Kimberly J; Pine, Daniel S; Blair, James; Wakschlag, Lauren S

2014-01-01

Youth and adults with psychopathic traits display disrupted reinforcement learning. Advances in measurement now enable examination of this association in preschoolers. The current study examines relations between reinforcement learning in preschoolers and parent ratings of reduced responsiveness to socialization, conceptualized as a developmental vulnerability to psychopathic traits. One hundred and fifty-seven preschoolers (mean age 4.7 ± 0.8 years) participated in a substudy that was embedded within a larger project. Children completed the 'Stars-in-Jars' task, which involved learning to select rewarded jars and avoid punished jars. Maternal report of responsiveness to socialization was assessed with the Punishment Insensitivity and Low Concern for Others scales of the Multidimensional Assessment of Preschool Disruptive Behavior (MAP-DB). Punishment Insensitivity, but not Low Concern for Others, was significantly associated with reinforcement learning in multivariate models that accounted for age and sex. Specifically, higher Punishment Insensitivity was associated with significantly lower overall performance and more errors on punished trials ('passive avoidance'). Impairments in reinforcement learning manifest in preschoolers who are high in maternal ratings of Punishment Insensitivity. If replicated, these findings may help to pinpoint the neurodevelopmental antecedents of psychopathic tendencies and suggest novel intervention targets beginning in early childhood. © 2013 The Authors. Journal of Child Psychology and Psychiatry © 2013 Association for Child and Adolescent Mental Health.

Effects of dopamine on reinforcement learning and consolidation in Parkinson's disease.

PubMed

Grogan, John P; Tsivos, Demitra; Smith, Laura; Knight, Brogan E; Bogacz, Rafal; Whone, Alan; Coulthard, Elizabeth J

2017-07-10

Emerging evidence suggests that dopamine may modulate learning and memory with important implications for understanding the neurobiology of memory and future therapeutic targeting. An influential hypothesis posits that dopamine biases reinforcement learning. More recent data also suggest an influence during both consolidation and retrieval. Eighteen Parkinson's disease patients learned through feedback ON or OFF medication, with memory tested 24 hr later ON or OFF medication (4 conditions, within-subjects design with matched healthy control group). Patients OFF medication during learning decreased in memory accuracy over the following 24 hr. In contrast to previous studies, however, dopaminergic medication during learning and testing did not affect expression of positive or negative reinforcement. Two further experiments were run without the 24 hr delay, but they too failed to reproduce effects of dopaminergic medication on reinforcement learning. While supportive of a dopaminergic role in consolidation, this study failed to replicate previous findings on reinforcement learning.

Autonomous Inter-Task Transfer in Reinforcement Learning Domains

DTIC Science & Technology

2008-08-01

Twentieth International Joint Conference on Artificial Intelli - gence, 2007. 304 Fumihide Tanaka and Masayuki Yamamura. Multitask reinforcement learning...Functions . . . . . . . . . . . . . . . . . . . . . . 17 2.2.3 Artificial Neural Networks . . . . . . . . . . . . . . . . . . . . 18 2.2.4 Instance-based...tures [Laird et al., 1986, Choi et al., 2007]. However, TL for RL tasks has only recently been gaining attention in the artificial intelligence

From Recurrent Choice to Skill Learning: A Reinforcement-Learning Model

ERIC Educational Resources Information Center

Fu, Wai-Tat; Anderson, John R.

2006-01-01

The authors propose a reinforcement-learning mechanism as a model for recurrent choice and extend it to account for skill learning. The model was inspired by recent research in neurophysiological studies of the basal ganglia and provides an integrated explanation of recurrent choice behavior and skill learning. The behavior includes effects of…

Implementation of real-time energy management strategy based on reinforcement learning for hybrid electric vehicles and simulation validation

PubMed Central

Kong, Zehui; Liu, Teng

2017-01-01

To further improve the fuel economy of series hybrid electric tracked vehicles, a reinforcement learning (RL)-based real-time energy management strategy is developed in this paper. In order to utilize the statistical characteristics of online driving schedule effectively, a recursive algorithm for the transition probability matrix (TPM) of power-request is derived. The reinforcement learning (RL) is applied to calculate and update the control policy at regular time, adapting to the varying driving conditions. A facing-forward powertrain model is built in detail, including the engine-generator model, battery model and vehicle dynamical model. The robustness and adaptability of real-time energy management strategy are validated through the comparison with the stationary control strategy based on initial transition probability matrix (TPM) generated from a long naturalistic driving cycle in the simulation. Results indicate that proposed method has better fuel economy than stationary one and is more effective in real-time control. PMID:28671967

Implementation of real-time energy management strategy based on reinforcement learning for hybrid electric vehicles and simulation validation.

PubMed

Kong, Zehui; Zou, Yuan; Liu, Teng

2017-01-01

To further improve the fuel economy of series hybrid electric tracked vehicles, a reinforcement learning (RL)-based real-time energy management strategy is developed in this paper. In order to utilize the statistical characteristics of online driving schedule effectively, a recursive algorithm for the transition probability matrix (TPM) of power-request is derived. The reinforcement learning (RL) is applied to calculate and update the control policy at regular time, adapting to the varying driving conditions. A facing-forward powertrain model is built in detail, including the engine-generator model, battery model and vehicle dynamical model. The robustness and adaptability of real-time energy management strategy are validated through the comparison with the stationary control strategy based on initial transition probability matrix (TPM) generated from a long naturalistic driving cycle in the simulation. Results indicate that proposed method has better fuel economy than stationary one and is more effective in real-time control.

Human reinforcement learning subdivides structured action spaces by learning effector-specific values

PubMed Central

Gershman, Samuel J.; Pesaran, Bijan; Daw, Nathaniel D.

2009-01-01

Humans and animals are endowed with a large number of effectors. Although this enables great behavioral flexibility, it presents an equally formidable reinforcement learning problem of discovering which actions are most valuable, due to the high dimensionality of the action space. An unresolved question is how neural systems for reinforcement learning – such as prediction error signals for action valuation associated with dopamine and the striatum – can cope with this “curse of dimensionality.” We propose a reinforcement learning framework that allows for learned action valuations to be decomposed into effector-specific components when appropriate to a task, and test it by studying to what extent human behavior and BOLD activity can exploit such a decomposition in a multieffector choice task. Subjects made simultaneous decisions with their left and right hands and received separate reward feedback for each hand movement. We found that choice behavior was better described by a learning model that decomposed the values of bimanual movements into separate values for each effector, rather than a traditional model that treated the bimanual actions as unitary with a single value. A decomposition of value into effector-specific components was also observed in value-related BOLD signaling, in the form of lateralized biases in striatal correlates of prediction error and anticipatory value correlates in the intraparietal sulcus. These results suggest that the human brain can use decomposed value representations to “divide and conquer” reinforcement learning over high-dimensional action spaces. PMID:19864565

Human reinforcement learning subdivides structured action spaces by learning effector-specific values.

PubMed

Gershman, Samuel J; Pesaran, Bijan; Daw, Nathaniel D

2009-10-28

Humans and animals are endowed with a large number of effectors. Although this enables great behavioral flexibility, it presents an equally formidable reinforcement learning problem of discovering which actions are most valuable because of the high dimensionality of the action space. An unresolved question is how neural systems for reinforcement learning-such as prediction error signals for action valuation associated with dopamine and the striatum-can cope with this "curse of dimensionality." We propose a reinforcement learning framework that allows for learned action valuations to be decomposed into effector-specific components when appropriate to a task, and test it by studying to what extent human behavior and blood oxygen level-dependent (BOLD) activity can exploit such a decomposition in a multieffector choice task. Subjects made simultaneous decisions with their left and right hands and received separate reward feedback for each hand movement. We found that choice behavior was better described by a learning model that decomposed the values of bimanual movements into separate values for each effector, rather than a traditional model that treated the bimanual actions as unitary with a single value. A decomposition of value into effector-specific components was also observed in value-related BOLD signaling, in the form of lateralized biases in striatal correlates of prediction error and anticipatory value correlates in the intraparietal sulcus. These results suggest that the human brain can use decomposed value representations to "divide and conquer" reinforcement learning over high-dimensional action spaces.

What Can Reinforcement Learning Teach Us About Non-Equilibrium Quantum Dynamics

NASA Astrophysics Data System (ADS)

Bukov, Marin; Day, Alexandre; Sels, Dries; Weinberg, Phillip; Polkovnikov, Anatoli; Mehta, Pankaj

Equilibrium thermodynamics and statistical physics are the building blocks of modern science and technology. Yet, our understanding of thermodynamic processes away from equilibrium is largely missing. In this talk, I will reveal the potential of what artificial intelligence can teach us about the complex behaviour of non-equilibrium systems. Specifically, I will discuss the problem of finding optimal drive protocols to prepare a desired target state in quantum mechanical systems by applying ideas from Reinforcement Learning [one can think of Reinforcement Learning as the study of how an agent (e.g. a robot) can learn and perfect a given policy through interactions with an environment.]. The driving protocols learnt by our agent suggest that the non-equilibrium world features possibilities easily defying intuition based on equilibrium physics.

A clustering-based graph Laplacian framework for value function approximation in reinforcement learning.

PubMed

Xu, Xin; Huang, Zhenhua; Graves, Daniel; Pedrycz, Witold

2014-12-01

In order to deal with the sequential decision problems with large or continuous state spaces, feature representation and function approximation have been a major research topic in reinforcement learning (RL). In this paper, a clustering-based graph Laplacian framework is presented for feature representation and value function approximation (VFA) in RL. By making use of clustering-based techniques, that is, K-means clustering or fuzzy C-means clustering, a graph Laplacian is constructed by subsampling in Markov decision processes (MDPs) with continuous state spaces. The basis functions for VFA can be automatically generated from spectral analysis of the graph Laplacian. The clustering-based graph Laplacian is integrated with a class of approximation policy iteration algorithms called representation policy iteration (RPI) for RL in MDPs with continuous state spaces. Simulation and experimental results show that, compared with previous RPI methods, the proposed approach needs fewer sample points to compute an efficient set of basis functions and the learning control performance can be improved for a variety of parameter settings.

The cerebellum: a neural system for the study of reinforcement learning.

PubMed

Swain, Rodney A; Kerr, Abigail L; Thompson, Richard F

2011-01-01

In its strictest application, the term "reinforcement learning" refers to a computational approach to learning in which an agent (often a machine) interacts with a mutable environment to maximize reward through trial and error. The approach borrows essentials from several fields, most notably Computer Science, Behavioral Neuroscience, and Psychology. At the most basic level, a neural system capable of mediating reinforcement learning must be able to acquire sensory information about the external environment and internal milieu (either directly or through connectivities with other brain regions), must be able to select a behavior to be executed, and must be capable of providing evaluative feedback about the success of that behavior. Given that Psychology informs us that reinforcers, both positive and negative, are stimuli or consequences that increase the probability that the immediately antecedent behavior will be repeated and that reinforcer strength or viability is modulated by the organism's past experience with the reinforcer, its affect, and even the state of its muscles (e.g., eyes open or closed); it is the case that any neural system that supports reinforcement learning must also be sensitive to these same considerations. Once learning is established, such a neural system must finally be able to maintain continued response expression and prevent response drift. In this report, we examine both historical and recent evidence that the cerebellum satisfies all of these requirements. While we report evidence from a variety of learning paradigms, the majority of our discussion will focus on classical conditioning of the rabbit eye blink response as an ideal model system for the study of reinforcement and reinforcement learning.

Preliminary Work for Examining the Scalability of Reinforcement Learning

NASA Technical Reports Server (NTRS)

Clouse, Jeff

1998-01-01

Researchers began studying automated agents that learn to perform multiple-step tasks early in the history of artificial intelligence (Samuel, 1963; Samuel, 1967; Waterman, 1970; Fikes, Hart & Nilsonn, 1972). Multiple-step tasks are tasks that can only be solved via a sequence of decisions, such as control problems, robotics problems, classic problem-solving, and game-playing. The objective of agents attempting to learn such tasks is to use the resources they have available in order to become more proficient at the tasks. In particular, each agent attempts to develop a good policy, a mapping from states to actions, that allows it to select actions that optimize a measure of its performance on the task; for example, reducing the number of steps necessary to complete the task successfully. Our study focuses on reinforcement learning, a set of learning techniques where the learner performs trial-and-error experiments in the task and adapts its policy based on the outcome of those experiments. Much of the work in reinforcement learning has focused on a particular, simple representation, where every problem state is represented explicitly in a table, and associated with each state are the actions that can be chosen in that state. A major advantage of this table lookup representation is that one can prove that certain reinforcement learning techniques will develop an optimal policy for the current task. The drawback is that the representation limits the application of reinforcement learning to multiple-step tasks with relatively small state-spaces. There has been a little theoretical work that proves that convergence to optimal solutions can be obtained when using generalization structures, but the structures are quite simple. The theory says little about complex structures, such as multi-layer, feedforward artificial neural networks (Rumelhart & McClelland, 1986), but empirical results indicate that the use of reinforcement learning with such structures is promising

When, What, and How Much to Reward in Reinforcement Learning-Based Models of Cognition

ERIC Educational Resources Information Center

Janssen, Christian P.; Gray, Wayne D.

2012-01-01

Reinforcement learning approaches to cognitive modeling represent task acquisition as learning to choose the sequence of steps that accomplishes the task while maximizing a reward. However, an apparently unrecognized problem for modelers is choosing when, what, and how much to reward; that is, when (the moment: end of trial, subtask, or some other…

A neural model of hierarchical reinforcement learning

PubMed Central

Rasmussen, Daniel; Eliasmith, Chris

2017-01-01

We develop a novel, biologically detailed neural model of reinforcement learning (RL) processes in the brain. This model incorporates a broad range of biological features that pose challenges to neural RL, such as temporally extended action sequences, continuous environments involving unknown time delays, and noisy/imprecise computations. Most significantly, we expand the model into the realm of hierarchical reinforcement learning (HRL), which divides the RL process into a hierarchy of actions at different levels of abstraction. Here we implement all the major components of HRL in a neural model that captures a variety of known anatomical and physiological properties of the brain. We demonstrate the performance of the model in a range of different environments, in order to emphasize the aim of understanding the brain’s general reinforcement learning ability. These results show that the model compares well to previous modelling work and demonstrates improved performance as a result of its hierarchical ability. We also show that the model’s behaviour is consistent with available data on human hierarchical RL, and generate several novel predictions. PMID:28683111

A neural model of hierarchical reinforcement learning.

PubMed

Rasmussen, Daniel; Voelker, Aaron; Eliasmith, Chris

2017-01-01

We develop a novel, biologically detailed neural model of reinforcement learning (RL) processes in the brain. This model incorporates a broad range of biological features that pose challenges to neural RL, such as temporally extended action sequences, continuous environments involving unknown time delays, and noisy/imprecise computations. Most significantly, we expand the model into the realm of hierarchical reinforcement learning (HRL), which divides the RL process into a hierarchy of actions at different levels of abstraction. Here we implement all the major components of HRL in a neural model that captures a variety of known anatomical and physiological properties of the brain. We demonstrate the performance of the model in a range of different environments, in order to emphasize the aim of understanding the brain's general reinforcement learning ability. These results show that the model compares well to previous modelling work and demonstrates improved performance as a result of its hierarchical ability. We also show that the model's behaviour is consistent with available data on human hierarchical RL, and generate several novel predictions.

Using Fuzzy Logic for Performance Evaluation in Reinforcement Learning

NASA Technical Reports Server (NTRS)

Berenji, Hamid R.; Khedkar, Pratap S.

1992-01-01

Current reinforcement learning algorithms require long training periods which generally limit their applicability to small size problems. A new architecture is described which uses fuzzy rules to initialize its two neural networks: a neural network for performance evaluation and another for action selection. This architecture is applied to control of dynamic systems and it is demonstrated that it is possible to start with an approximate prior knowledge and learn to refine it through experiments using reinforcement learning.

An Upside to Reward Sensitivity: The Hippocampus Supports Enhanced Reinforcement Learning in Adolescence.

PubMed

Davidow, Juliet Y; Foerde, Karin; Galván, Adriana; Shohamy, Daphna

2016-10-05

Adolescents are notorious for engaging in reward-seeking behaviors, a tendency attributed to heightened activity in the brain's reward systems during adolescence. It has been suggested that reward sensitivity in adolescence might be adaptive, but evidence of an adaptive role has been scarce. Using a probabilistic reinforcement learning task combined with reinforcement learning models and fMRI, we found that adolescents showed better reinforcement learning and a stronger link between reinforcement learning and episodic memory for rewarding outcomes. This behavioral benefit was related to heightened prediction error-related BOLD activity in the hippocampus and to stronger functional connectivity between the hippocampus and the striatum at the time of reinforcement. These findings reveal an important role for the hippocampus in reinforcement learning in adolescence and suggest that reward sensitivity in adolescence is related to adaptive differences in how adolescents learn from experience. Copyright © 2016 Elsevier Inc. All rights reserved.

Design issues for a reinforcement-based self-learning fuzzy controller

NASA Technical Reports Server (NTRS)

Yen, John; Wang, Haojin; Dauherity, Walter

1993-01-01

Fuzzy logic controllers have some often cited advantages over conventional techniques such as PID control: easy implementation, its accommodation to natural language, the ability to cover wider range of operating conditions and others. One major obstacle that hinders its broader application is the lack of a systematic way to develop and modify its rules and as result the creation and modification of fuzzy rules often depends on try-error or pure experimentation. One of the proposed approaches to address this issue is self-learning fuzzy logic controllers (SFLC) that use reinforcement learning techniques to learn the desirability of states and to adjust the consequent part of fuzzy control rules accordingly. Due to the different dynamics of the controlled processes, the performance of self-learning fuzzy controller is highly contingent on the design. The design issue has not received sufficient attention. The issues related to the design of a SFLC for the application to chemical process are discussed and its performance is compared with that of PID and self-tuning fuzzy logic controller.

Role of dopamine D2 receptors in human reinforcement learning.

PubMed

Eisenegger, Christoph; Naef, Michael; Linssen, Anke; Clark, Luke; Gandamaneni, Praveen K; Müller, Ulrich; Robbins, Trevor W

2014-09-01

Influential neurocomputational models emphasize dopamine (DA) as an electrophysiological and neurochemical correlate of reinforcement learning. However, evidence of a specific causal role of DA receptors in learning has been less forthcoming, especially in humans. Here we combine, in a between-subjects design, administration of a high dose of the selective DA D2/3-receptor antagonist sulpiride with genetic analysis of the DA D2 receptor in a behavioral study of reinforcement learning in a sample of 78 healthy male volunteers. In contrast to predictions of prevailing models emphasizing DA's pivotal role in learning via prediction errors, we found that sulpiride did not disrupt learning, but rather induced profound impairments in choice performance. The disruption was selective for stimuli indicating reward, whereas loss avoidance performance was unaffected. Effects were driven by volunteers with higher serum levels of the drug, and in those with genetically determined lower density of striatal DA D2 receptors. This is the clearest demonstration to date for a causal modulatory role of the DA D2 receptor in choice performance that might be distinct from learning. Our findings challenge current reward prediction error models of reinforcement learning, and suggest that classical animal models emphasizing a role of postsynaptic DA D2 receptors in motivational aspects of reinforcement learning may apply to humans as well.

Role of Dopamine D2 Receptors in Human Reinforcement Learning

PubMed Central

Eisenegger, Christoph; Naef, Michael; Linssen, Anke; Clark, Luke; Gandamaneni, Praveen K; Müller, Ulrich; Robbins, Trevor W

2014-01-01

Influential neurocomputational models emphasize dopamine (DA) as an electrophysiological and neurochemical correlate of reinforcement learning. However, evidence of a specific causal role of DA receptors in learning has been less forthcoming, especially in humans. Here we combine, in a between-subjects design, administration of a high dose of the selective DA D2/3-receptor antagonist sulpiride with genetic analysis of the DA D2 receptor in a behavioral study of reinforcement learning in a sample of 78 healthy male volunteers. In contrast to predictions of prevailing models emphasizing DA's pivotal role in learning via prediction errors, we found that sulpiride did not disrupt learning, but rather induced profound impairments in choice performance. The disruption was selective for stimuli indicating reward, whereas loss avoidance performance was unaffected. Effects were driven by volunteers with higher serum levels of the drug, and in those with genetically determined lower density of striatal DA D2 receptors. This is the clearest demonstration to date for a causal modulatory role of the DA D2 receptor in choice performance that might be distinct from learning. Our findings challenge current reward prediction error models of reinforcement learning, and suggest that classical animal models emphasizing a role of postsynaptic DA D2 receptors in motivational aspects of reinforcement learning may apply to humans as well. PMID:24713613

Intelligence moderates reinforcement learning: a mini-review of the neural evidence

PubMed Central

2014-01-01

Our understanding of the neural basis of reinforcement learning and intelligence, two key factors contributing to human strivings, has progressed significantly recently. However, the overlap of these two lines of research, namely, how intelligence affects neural responses during reinforcement learning, remains uninvestigated. A mini-review of three existing studies suggests that higher IQ (especially fluid IQ) may enhance the neural signal of positive prediction error in dorsolateral prefrontal cortex, dorsal anterior cingulate cortex, and striatum, several brain substrates of reinforcement learning or intelligence. PMID:25185818

Refining Linear Fuzzy Rules by Reinforcement Learning

NASA Technical Reports Server (NTRS)

Berenji, Hamid R.; Khedkar, Pratap S.; Malkani, Anil

1996-01-01

Linear fuzzy rules are increasingly being used in the development of fuzzy logic systems. Radial basis functions have also been used in the antecedents of the rules for clustering in product space which can automatically generate a set of linear fuzzy rules from an input/output data set. Manual methods are usually used in refining these rules. This paper presents a method for refining the parameters of these rules using reinforcement learning which can be applied in domains where supervised input-output data is not available and reinforcements are received only after a long sequence of actions. This is shown for a generalization of radial basis functions. The formation of fuzzy rules from data and their automatic refinement is an important step in closing the gap between the application of reinforcement learning methods in the domains where only some limited input-output data is available.

Reinforcement Learning Strategies for Clinical Trials in Non-small Cell Lung Cancer

PubMed Central

Zhao, Yufan; Zeng, Donglin; Socinski, Mark A.; Kosorok, Michael R.

2010-01-01

Summary Typical regimens for advanced metastatic stage IIIB/IV non-small cell lung cancer (NSCLC) consist of multiple lines of treatment. We present an adaptive reinforcement learning approach to discover optimal individualized treatment regimens from a specially designed clinical trial (a “clinical reinforcement trial”) of an experimental treatment for patients with advanced NSCLC who have not been treated previously with systemic therapy. In addition to the complexity of the problem of selecting optimal compounds for first and second-line treatments based on prognostic factors, another primary goal is to determine the optimal time to initiate second-line therapy, either immediately or delayed after induction therapy, yielding the longest overall survival time. A reinforcement learning method called Q-learning is utilized which involves learning an optimal regimen from patient data generated from the clinical reinforcement trial. Approximating the Q-function with time-indexed parameters can be achieved by using a modification of support vector regression which can utilize censored data. Within this framework, a simulation study shows that the procedure can extract optimal regimens for two lines of treatment directly from clinical data without prior knowledge of the treatment effect mechanism. In addition, we demonstrate that the design reliably selects the best initial time for second-line therapy while taking into account the heterogeneity of NSCLC across patients. PMID:21385164

Reinforcement of Science Learning through Local Culture: A Delphi Study

ERIC Educational Resources Information Center

Nuangchalerm, Prasart

2008-01-01

This study aims to explore the ways to reinforce science learning through local culture by using Delphi technique. Twenty four participants in various fields of study were selected. The result of study provides a framework for reinforcement of science learning through local culture on the theme life and environment. (Contains 1 table.)

Intelligence moderates reinforcement learning: a mini-review of the neural evidence.

PubMed

Chen, Chong

2015-06-01

Our understanding of the neural basis of reinforcement learning and intelligence, two key factors contributing to human strivings, has progressed significantly recently. However, the overlap of these two lines of research, namely, how intelligence affects neural responses during reinforcement learning, remains uninvestigated. A mini-review of three existing studies suggests that higher IQ (especially fluid IQ) may enhance the neural signal of positive prediction error in dorsolateral prefrontal cortex, dorsal anterior cingulate cortex, and striatum, several brain substrates of reinforcement learning or intelligence. Copyright © 2015 the American Physiological Society.

Feature Reinforcement Learning: Part I. Unstructured MDPs

NASA Astrophysics Data System (ADS)

Hutter, Marcus

2009-12-01

General-purpose, intelligent, learning agents cycle through sequences of observations, actions, and rewards that are complex, uncertain, unknown, and non-Markovian. On the other hand, reinforcement learning is well-developed for small finite state Markov decision processes (MDPs). Up to now, extracting the right state representations out of bare observations, that is, reducing the general agent setup to the MDP framework, is an art that involves significant effort by designers. The primary goal of this work is to automate the reduction process and thereby significantly expand the scope of many existing reinforcement learning algorithms and the agents that employ them. Before we can think of mechanizing this search for suitable MDPs, we need a formal objective criterion. The main contribution of this article is to develop such a criterion. I also integrate the various parts into one learning algorithm. Extensions to more realistic dynamic Bayesian networks are developed in Part II (Hutter, 2009c). The role of POMDPs is also considered there.

A Discussion of Possibility of Reinforcement Learning Using Event-Related Potential in BCI

NASA Astrophysics Data System (ADS)

Yamagishi, Yuya; Tsubone, Tadashi; Wada, Yasuhiro

Recently, Brain computer interface (BCI) which is a direct connecting pathway an external device such as a computer or a robot and a human brain have gotten a lot of attention. Since BCI can control the machines as robots by using the brain activity without using the voluntary muscle, the BCI may become a useful communication tool for handicapped persons, for instance, amyotrophic lateral sclerosis patients. However, in order to realize the BCI system which can perform precise tasks on various environments, it is necessary to design the control rules to adapt to the dynamic environments. Reinforcement learning is one approach of the design of the control rule. If this reinforcement leaning can be performed by the brain activity, it leads to the attainment of BCI that has general versatility. In this research, we paid attention to P300 of event-related potential as an alternative signal of the reward of reinforcement learning. We discriminated between the success and the failure trials from P300 of the EEG of the single trial by using the proposed discrimination algorithm based on Support vector machine. The possibility of reinforcement learning was examined from the viewpoint of the number of discriminated trials. It was shown that there was a possibility to be able to learn in most subjects.

Effects of dopamine on reinforcement learning and consolidation in Parkinson’s disease

PubMed Central

Grogan, John P; Tsivos, Demitra; Smith, Laura; Knight, Brogan E; Bogacz, Rafal; Whone, Alan; Coulthard, Elizabeth J

2017-01-01

Emerging evidence suggests that dopamine may modulate learning and memory with important implications for understanding the neurobiology of memory and future therapeutic targeting. An influential hypothesis posits that dopamine biases reinforcement learning. More recent data also suggest an influence during both consolidation and retrieval. Eighteen Parkinson’s disease patients learned through feedback ON or OFF medication, with memory tested 24 hr later ON or OFF medication (4 conditions, within-subjects design with matched healthy control group). Patients OFF medication during learning decreased in memory accuracy over the following 24 hr. In contrast to previous studies, however, dopaminergic medication during learning and testing did not affect expression of positive or negative reinforcement. Two further experiments were run without the 24 hr delay, but they too failed to reproduce effects of dopaminergic medication on reinforcement learning. While supportive of a dopaminergic role in consolidation, this study failed to replicate previous findings on reinforcement learning. DOI: http://dx.doi.org/10.7554/eLife.26801.001 PMID:28691905

"Notice of Violation of IEEE Publication Principles" Multiobjective Reinforcement Learning: A Comprehensive Overview.

PubMed

Liu, Chunming; Xu, Xin; Hu, Dewen

2013-04-29

Reinforcement learning is a powerful mechanism for enabling agents to learn in an unknown environment, and most reinforcement learning algorithms aim to maximize some numerical value, which represents only one long-term objective. However, multiple long-term objectives are exhibited in many real-world decision and control problems; therefore, recently, there has been growing interest in solving multiobjective reinforcement learning (MORL) problems with multiple conflicting objectives. The aim of this paper is to present a comprehensive overview of MORL. In this paper, the basic architecture, research topics, and naive solutions of MORL are introduced at first. Then, several representative MORL approaches and some important directions of recent research are reviewed. The relationships between MORL and other related research are also discussed, which include multiobjective optimization, hierarchical reinforcement learning, and multi-agent reinforcement learning. Finally, research challenges and open problems of MORL techniques are highlighted.

Learning and altering behaviours by reinforcement: neurocognitive differences between children and adults.

PubMed

Shephard, E; Jackson, G M; Groom, M J

2014-01-01

This study examined neurocognitive differences between children and adults in the ability to learn and adapt simple stimulus-response associations through feedback. Fourteen typically developing children (mean age=10.2) and 15 healthy adults (mean age=25.5) completed a simple task in which they learned to associate visually presented stimuli with manual responses based on performance feedback (acquisition phase), and then reversed and re-learned those associations following an unexpected change in reinforcement contingencies (reversal phase). Electrophysiological activity was recorded throughout task performance. We found no group differences in learning-related changes in performance (reaction time, accuracy) or in the amplitude of event-related potentials (ERPs) associated with stimulus processing (P3 ERP) or feedback processing (feedback-related negativity; FRN) during the acquisition phase. However, children's performance was significantly more disrupted by the reversal than adults and FRN amplitudes were significantly modulated by the reversal phase in children but not adults. These findings indicate that children have specific difficulties with reinforcement learning when acquired behaviours must be altered. This may be caused by the added demands on immature executive functioning, specifically response monitoring, created by the requirement to reverse the associations, or a developmental difference in the way in which children and adults approach reinforcement learning. Copyright © 2013 The Authors. Published by Elsevier Ltd.. All rights reserved.

Online human training of a myoelectric prosthesis controller via actor-critic reinforcement learning.

PubMed

Pilarski, Patrick M; Dawson, Michael R; Degris, Thomas; Fahimi, Farbod; Carey, Jason P; Sutton, Richard S

2011-01-01

As a contribution toward the goal of adaptable, intelligent artificial limbs, this work introduces a continuous actor-critic reinforcement learning method for optimizing the control of multi-function myoelectric devices. Using a simulated upper-arm robotic prosthesis, we demonstrate how it is possible to derive successful limb controllers from myoelectric data using only a sparse human-delivered training signal, without requiring detailed knowledge about the task domain. This reinforcement-based machine learning framework is well suited for use by both patients and clinical staff, and may be easily adapted to different application domains and the needs of individual amputees. To our knowledge, this is the first my-oelectric control approach that facilitates the online learning of new amputee-specific motions based only on a one-dimensional (scalar) feedback signal provided by the user of the prosthesis. © 2011 IEEE

Antipsychotic dose modulates behavioral and neural responses to feedback during reinforcement learning in schizophrenia.

PubMed

Insel, Catherine; Reinen, Jenna; Weber, Jochen; Wager, Tor D; Jarskog, L Fredrik; Shohamy, Daphna; Smith, Edward E

2014-03-01

Schizophrenia is characterized by an abnormal dopamine system, and dopamine blockade is the primary mechanism of antipsychotic treatment. Consistent with the known role of dopamine in reward processing, prior research has demonstrated that patients with schizophrenia exhibit impairments in reward-based learning. However, it remains unknown how treatment with antipsychotic medication impacts the behavioral and neural signatures of reinforcement learning in schizophrenia. The goal of this study was to examine whether antipsychotic medication modulates behavioral and neural responses to prediction error coding during reinforcement learning. Patients with schizophrenia completed a reinforcement learning task while undergoing functional magnetic resonance imaging. The task consisted of two separate conditions in which participants accumulated monetary gain or avoided monetary loss. Behavioral results indicated that antipsychotic medication dose was associated with altered behavioral approaches to learning, such that patients taking higher doses of medication showed increased sensitivity to negative reinforcement. Higher doses of antipsychotic medication were also associated with higher learning rates (LRs), suggesting that medication enhanced sensitivity to trial-by-trial feedback. Neuroimaging data demonstrated that antipsychotic dose was related to differences in neural signatures of feedback prediction error during the loss condition. Specifically, patients taking higher doses of medication showed attenuated prediction error responses in the striatum and the medial prefrontal cortex. These findings indicate that antipsychotic medication treatment may influence motivational processes in patients with schizophrenia.

Longitudinal investigation on learned helplessness tested under negative and positive reinforcement involving stimulus control.

PubMed

Oliveira, Emileane C; Hunziker, Maria Helena

2014-07-01

In this study, we investigated whether (a) animals demonstrating the learned helplessness effect during an escape contingency also show learning deficits under positive reinforcement contingencies involving stimulus control and (b) the exposure to positive reinforcement contingencies eliminates the learned helplessness effect under an escape contingency. Rats were initially exposed to controllable (C), uncontrollable (U) or no (N) shocks. After 24h, they were exposed to 60 escapable shocks delivered in a shuttlebox. In the following phase, we selected from each group the four subjects that presented the most typical group pattern: no escape learning (learned helplessness effect) in Group U and escape learning in Groups C and N. All subjects were then exposed to two phases, the (1) positive reinforcement for lever pressing under a multiple FR/Extinction schedule and (2) a re-test under negative reinforcement (escape). A fourth group (n=4) was exposed only to the positive reinforcement sessions. All subjects showed discrimination learning under multiple schedule. In the escape re-test, the learned helplessness effect was maintained for three of the animals in Group U. These results suggest that the learned helplessness effect did not extend to discriminative behavior that is positively reinforced and that the learned helplessness effect did not revert for most subjects after exposure to positive reinforcement. We discuss some theoretical implications as related to learned helplessness as an effect restricted to aversive contingencies and to the absence of reversion after positive reinforcement. This article is part of a Special Issue entitled: insert SI title. Copyright © 2014. Published by Elsevier B.V.

Microstimulation of the Human Substantia Nigra Alters Reinforcement Learning

PubMed Central

Ramayya, Ashwin G.; Misra, Amrit

2014-01-01

Animal studies have shown that substantia nigra (SN) dopaminergic (DA) neurons strengthen action–reward associations during reinforcement learning, but their role in human learning is not known. Here, we applied microstimulation in the SN of 11 patients undergoing deep brain stimulation surgery for the treatment of Parkinson's disease as they performed a two-alternative probability learning task in which rewards were contingent on stimuli, rather than actions. Subjects demonstrated decreased learning from reward trials that were accompanied by phasic SN microstimulation compared with reward trials without stimulation. Subjects who showed large decreases in learning also showed an increased bias toward repeating actions after stimulation trials; therefore, stimulation may have decreased learning by strengthening action–reward associations rather than stimulus–reward associations. Our findings build on previous studies implicating SN DA neurons in preferentially strengthening action–reward associations during reinforcement learning. PMID:24828643

The Role of Multiple Neuromodulators in Reinforcement Learning That Is Based on Competition between Eligibility Traces.

PubMed

Huertas, Marco A; Schwettmann, Sarah E; Shouval, Harel Z

2016-01-01

The ability to maximize reward and avoid punishment is essential for animal survival. Reinforcement learning (RL) refers to the algorithms used by biological or artificial systems to learn how to maximize reward or avoid negative outcomes based on past experiences. While RL is also important in machine learning, the types of mechanistic constraints encountered by biological machinery might be different than those for artificial systems. Two major problems encountered by RL are how to relate a stimulus with a reinforcing signal that is delayed in time (temporal credit assignment), and how to stop learning once the target behaviors are attained (stopping rule). To address the first problem synaptic eligibility traces were introduced, bridging the temporal gap between a stimulus and its reward. Although, these were mere theoretical constructs, recent experiments have provided evidence of their existence. These experiments also reveal that the presence of specific neuromodulators converts the traces into changes in synaptic efficacy. A mechanistic implementation of the stopping rule usually assumes the inhibition of the reward nucleus; however, recent experimental results have shown that learning terminates at the appropriate network state even in setups where the reward nucleus cannot be inhibited. In an effort to describe a learning rule that solves the temporal credit assignment problem and implements a biologically plausible stopping rule, we proposed a model based on two separate synaptic eligibility traces, one for long-term potentiation (LTP) and one for long-term depression (LTD), each obeying different dynamics and having different effective magnitudes. The model has been shown to successfully generate stable learning in recurrent networks. Although, the model assumes the presence of a single neuromodulator, evidence indicates that there are different neuromodulators for expressing the different traces. What could be the role of different neuromodulators for

Optimal and Autonomous Control Using Reinforcement Learning: A Survey.

PubMed

Kiumarsi, Bahare; Vamvoudakis, Kyriakos G; Modares, Hamidreza; Lewis, Frank L

2018-06-01

This paper reviews the current state of the art on reinforcement learning (RL)-based feedback control solutions to optimal regulation and tracking of single and multiagent systems. Existing RL solutions to both optimal and control problems, as well as graphical games, will be reviewed. RL methods learn the solution to optimal control and game problems online and using measured data along the system trajectories. We discuss Q-learning and the integral RL algorithm as core algorithms for discrete-time (DT) and continuous-time (CT) systems, respectively. Moreover, we discuss a new direction of off-policy RL for both CT and DT systems. Finally, we review several applications.

Adaptive Educational Software by Applying Reinforcement Learning

ERIC Educational Resources Information Center

Bennane, Abdellah

2013-01-01

The introduction of the intelligence in teaching software is the object of this paper. In software elaboration process, one uses some learning techniques in order to adapt the teaching software to characteristics of student. Generally, one uses the artificial intelligence techniques like reinforcement learning, Bayesian network in order to adapt…

Investigation of a Reinforcement-Based Toilet Training Procedure for Children with Autism.

ERIC Educational Resources Information Center

Cicero, Frank R.; Pfadt, Al

2002-01-01

This study evaluated the effectiveness of a reinforcement-based toilet training intervention with three children with autism. Procedures included positive reinforcement, graduated guidance, scheduled practice trials, and forward prompting. All three children reduced urination accidents to zero and learned to request bathroom use spontaneously…

How we learn to make decisions: rapid propagation of reinforcement learning prediction errors in humans.

PubMed

Krigolson, Olav E; Hassall, Cameron D; Handy, Todd C

2014-03-01

Our ability to make decisions is predicated upon our knowledge of the outcomes of the actions available to us. Reinforcement learning theory posits that actions followed by a reward or punishment acquire value through the computation of prediction errors-discrepancies between the predicted and the actual reward. A multitude of neuroimaging studies have demonstrated that rewards and punishments evoke neural responses that appear to reflect reinforcement learning prediction errors [e.g., Krigolson, O. E., Pierce, L. J., Holroyd, C. B., & Tanaka, J. W. Learning to become an expert: Reinforcement learning and the acquisition of perceptual expertise. Journal of Cognitive Neuroscience, 21, 1833-1840, 2009; Bayer, H. M., & Glimcher, P. W. Midbrain dopamine neurons encode a quantitative reward prediction error signal. Neuron, 47, 129-141, 2005; O'Doherty, J. P. Reward representations and reward-related learning in the human brain: Insights from neuroimaging. Current Opinion in Neurobiology, 14, 769-776, 2004; Holroyd, C. B., & Coles, M. G. H. The neural basis of human error processing: Reinforcement learning, dopamine, and the error-related negativity. Psychological Review, 109, 679-709, 2002]. Here, we used the brain ERP technique to demonstrate that not only do rewards elicit a neural response akin to a prediction error but also that this signal rapidly diminished and propagated to the time of choice presentation with learning. Specifically, in a simple, learnable gambling task, we show that novel rewards elicited a feedback error-related negativity that rapidly decreased in amplitude with learning. Furthermore, we demonstrate the existence of a reward positivity at choice presentation, a previously unreported ERP component that has a similar timing and topography as the feedback error-related negativity that increased in amplitude with learning. The pattern of results we observed mirrored the output of a computational model that we implemented to compute reward

Reinforcement learning in professional basketball players

PubMed Central

Neiman, Tal; Loewenstein, Yonatan

2011-01-01

Reinforcement learning in complex natural environments is a challenging task because the agent should generalize from the outcomes of actions taken in one state of the world to future actions in different states of the world. The extent to which human experts find the proper level of generalization is unclear. Here we show, using the sequences of field goal attempts made by professional basketball players, that the outcome of even a single field goal attempt has a considerable effect on the rate of subsequent 3 point shot attempts, in line with standard models of reinforcement learning. However, this change in behaviour is associated with negative correlations between the outcomes of successive field goal attempts. These results indicate that despite years of experience and high motivation, professional players overgeneralize from the outcomes of their most recent actions, which leads to decreased performance. PMID:22146388

Flow Navigation by Smart Microswimmers via Reinforcement Learning

NASA Astrophysics Data System (ADS)

Colabrese, Simona; Biferale, Luca; Celani, Antonio; Gustavsson, Kristian

2017-11-01

We have numerically modeled active particles which are able to acquire some limited knowledge of the fluid environment from simple mechanical cues and exert a control on their preferred steering direction. We show that those swimmers can learn effective strategies just by experience, using a reinforcement learning algorithm. As an example, we focus on smart gravitactic swimmers. These are active particles whose task is to reach the highest altitude within some time horizon, exploiting the underlying flow whenever possible. The reinforcement learning algorithm allows particles to learn effective strategies even in difficult situations when, in the absence of control, they would end up being trapped by flow structures. These strategies are highly nontrivial and cannot be easily guessed in advance. This work paves the way towards the engineering of smart microswimmers that solve difficult navigation problems. ERC AdG NewTURB 339032.

Dissociable neural representations of reinforcement and belief prediction errors underlie strategic learning

PubMed Central

Zhu, Lusha; Mathewson, Kyle E.; Hsu, Ming

2012-01-01

Decision-making in the presence of other competitive intelligent agents is fundamental for social and economic behavior. Such decisions require agents to behave strategically, where in addition to learning about the rewards and punishments available in the environment, they also need to anticipate and respond to actions of others competing for the same rewards. However, whereas we know much about strategic learning at both theoretical and behavioral levels, we know relatively little about the underlying neural mechanisms. Here, we show using a multi-strategy competitive learning paradigm that strategic choices can be characterized by extending the reinforcement learning (RL) framework to incorporate agents’ beliefs about the actions of their opponents. Furthermore, using this characterization to generate putative internal values, we used model-based functional magnetic resonance imaging to investigate neural computations underlying strategic learning. We found that the distinct notions of prediction errors derived from our computational model are processed in a partially overlapping but distinct set of brain regions. Specifically, we found that the RL prediction error was correlated with activity in the ventral striatum. In contrast, activity in the ventral striatum, as well as the rostral anterior cingulate (rACC), was correlated with a previously uncharacterized belief-based prediction error. Furthermore, activity in rACC reflected individual differences in degree of engagement in belief learning. These results suggest a model of strategic behavior where learning arises from interaction of dissociable reinforcement and belief-based inputs. PMID:22307594

Dissociable neural representations of reinforcement and belief prediction errors underlie strategic learning.

PubMed

Zhu, Lusha; Mathewson, Kyle E; Hsu, Ming

2012-01-31

Decision-making in the presence of other competitive intelligent agents is fundamental for social and economic behavior. Such decisions require agents to behave strategically, where in addition to learning about the rewards and punishments available in the environment, they also need to anticipate and respond to actions of others competing for the same rewards. However, whereas we know much about strategic learning at both theoretical and behavioral levels, we know relatively little about the underlying neural mechanisms. Here, we show using a multi-strategy competitive learning paradigm that strategic choices can be characterized by extending the reinforcement learning (RL) framework to incorporate agents' beliefs about the actions of their opponents. Furthermore, using this characterization to generate putative internal values, we used model-based functional magnetic resonance imaging to investigate neural computations underlying strategic learning. We found that the distinct notions of prediction errors derived from our computational model are processed in a partially overlapping but distinct set of brain regions. Specifically, we found that the RL prediction error was correlated with activity in the ventral striatum. In contrast, activity in the ventral striatum, as well as the rostral anterior cingulate (rACC), was correlated with a previously uncharacterized belief-based prediction error. Furthermore, activity in rACC reflected individual differences in degree of engagement in belief learning. These results suggest a model of strategic behavior where learning arises from interaction of dissociable reinforcement and belief-based inputs.

Flow Navigation by Smart Microswimmers via Reinforcement Learning

NASA Astrophysics Data System (ADS)

Colabrese, Simona; Gustavsson, Kristian; Celani, Antonio; Biferale, Luca

2017-04-01

Smart active particles can acquire some limited knowledge of the fluid environment from simple mechanical cues and exert a control on their preferred steering direction. Their goal is to learn the best way to navigate by exploiting the underlying flow whenever possible. As an example, we focus our attention on smart gravitactic swimmers. These are active particles whose task is to reach the highest altitude within some time horizon, given the constraints enforced by fluid mechanics. By means of numerical experiments, we show that swimmers indeed learn nearly optimal strategies just by experience. A reinforcement learning algorithm allows particles to learn effective strategies even in difficult situations when, in the absence of control, they would end up being trapped by flow structures. These strategies are highly nontrivial and cannot be easily guessed in advance. This Letter illustrates the potential of reinforcement learning algorithms to model adaptive behavior in complex flows and paves the way towards the engineering of smart microswimmers that solve difficult navigation problems.

Optimal control in microgrid using multi-agent reinforcement learning.

PubMed

Li, Fu-Dong; Wu, Min; He, Yong; Chen, Xin

2012-11-01

This paper presents an improved reinforcement learning method to minimize electricity costs on the premise of satisfying the power balance and generation limit of units in a microgrid with grid-connected mode. Firstly, the microgrid control requirements are analyzed and the objective function of optimal control for microgrid is proposed. Then, a state variable "Average Electricity Price Trend" which is used to express the most possible transitions of the system is developed so as to reduce the complexity and randomicity of the microgrid, and a multi-agent architecture including agents, state variables, action variables and reward function is formulated. Furthermore, dynamic hierarchical reinforcement learning, based on change rate of key state variable, is established to carry out optimal policy exploration. The analysis shows that the proposed method is beneficial to handle the problem of "curse of dimensionality" and speed up learning in the unknown large-scale world. Finally, the simulation results under JADE (Java Agent Development Framework) demonstrate the validity of the presented method in optimal control for a microgrid with grid-connected mode. Copyright © 2012 ISA. Published by Elsevier Ltd. All rights reserved.

Application of fuzzy logic-neural network based reinforcement learning to proximity and docking operations: Special approach/docking testcase results

NASA Technical Reports Server (NTRS)

Jani, Yashvant

1993-01-01

As part of the RICIS project, the reinforcement learning techniques developed at Ames Research Center are being applied to proximity and docking operations using the Shuttle and Solar Maximum Mission (SMM) satellite simulation. In utilizing these fuzzy learning techniques, we use the Approximate Reasoning based Intelligent Control (ARIC) architecture, and so we use these two terms interchangeably to imply the same. This activity is carried out in the Software Technology Laboratory utilizing the Orbital Operations Simulator (OOS) and programming/testing support from other contractor personnel. This report is the final deliverable D4 in our milestones and project activity. It provides the test results for the special testcase of approach/docking scenario for the shuttle and SMM satellite. Based on our experience and analysis with the attitude and translational controllers, we have modified the basic configuration of the reinforcement learning algorithm in ARIC. The shuttle translational controller and its implementation in ARIC is described in our deliverable D3. In order to simulate the final approach and docking operations, we have set-up this special testcase as described in section 2. The ARIC performance results for these operations are discussed in section 3 and conclusions are provided in section 4 along with the summary for the project.

Embedded Incremental Feature Selection for Reinforcement Learning

DTIC Science & Technology

2012-05-01

Prior to this work, feature selection for reinforce- ment learning has focused on linear value function ap- proximation ( Kolter and Ng, 2009; Parr et al...InProceed- ings of the the 23rd International Conference on Ma- chine Learning, pages 449–456. Kolter , J. Z. and Ng, A. Y. (2009). Regularization and feature

Reinforcement learning with Marr.

PubMed

Niv, Yael; Langdon, Angela

2016-10-01

To many, the poster child for David Marr's famous three levels of scientific inquiry is reinforcement learning-a computational theory of reward optimization, which readily prescribes algorithmic solutions that evidence striking resemblance to signals found in the brain, suggesting a straightforward neural implementation. Here we review questions that remain open at each level of analysis, concluding that the path forward to their resolution calls for inspiration across levels, rather than a focus on mutual constraints.

Universal effect of dynamical reinforcement learning mechanism in spatial evolutionary games

NASA Astrophysics Data System (ADS)

Zhang, Hai-Feng; Wu, Zhi-Xi; Wang, Bing-Hong

2012-06-01

One of the prototypical mechanisms in understanding the ubiquitous cooperation in social dilemma situations is the win-stay, lose-shift rule. In this work, a generalized win-stay, lose-shift learning model—a reinforcement learning model with dynamic aspiration level—is proposed to describe how humans adapt their social behaviors based on their social experiences. In the model, the players incorporate the information of the outcomes in previous rounds with time-dependent aspiration payoffs to regulate the probability of choosing cooperation. By investigating such a reinforcement learning rule in the spatial prisoner's dilemma game and public goods game, a most noteworthy viewpoint is that moderate greediness (i.e. moderate aspiration level) favors best the development and organization of collective cooperation. The generality of this observation is tested against different regulation strengths and different types of network of interaction as well. We also make comparisons with two recently proposed models to highlight the importance of the mechanism of adaptive aspiration level in supporting cooperation in structured populations.

Reinforcement learning: Solving two case studies

NASA Astrophysics Data System (ADS)

Duarte, Ana Filipa; Silva, Pedro; dos Santos, Cristina Peixoto

2012-09-01

Reinforcement Learning algorithms offer interesting features for the control of autonomous systems, such as the ability to learn from direct interaction with the environment, and the use of a simple reward signalas opposed to the input-outputs pairsused in classic supervised learning. The reward signal indicates the success of failure of the actions executed by the agent in the environment. In this work, are described RL algorithmsapplied to two case studies: the Crawler robot and the widely known inverted pendulum. We explore RL capabilities to autonomously learn a basic locomotion pattern in the Crawler, andapproach the balancing problem of biped locomotion using the inverted pendulum.

Oculomotor learning revisited: a model of reinforcement learning in the basal ganglia incorporating an efference copy of motor actions

PubMed Central

Fee, Michale S.

2012-01-01

In its simplest formulation, reinforcement learning is based on the idea that if an action taken in a particular context is followed by a favorable outcome, then, in the same context, the tendency to produce that action should be strengthened, or reinforced. While reinforcement learning forms the basis of many current theories of basal ganglia (BG) function, these models do not incorporate distinct computational roles for signals that convey context, and those that convey what action an animal takes. Recent experiments in the songbird suggest that vocal-related BG circuitry receives two functionally distinct excitatory inputs. One input is from a cortical region that carries context information about the current “time” in the motor sequence. The other is an efference copy of motor commands from a separate cortical brain region that generates vocal variability during learning. Based on these findings, I propose here a general model of vertebrate BG function that combines context information with a distinct motor efference copy signal. The signals are integrated by a learning rule in which efference copy inputs gate the potentiation of context inputs (but not efference copy inputs) onto medium spiny neurons in response to a rewarded action. The hypothesis is described in terms of a circuit that implements the learning of visually guided saccades. The model makes testable predictions about the anatomical and functional properties of hypothesized context and efference copy inputs to the striatum from both thalamic and cortical sources. PMID:22754501

Oculomotor learning revisited: a model of reinforcement learning in the basal ganglia incorporating an efference copy of motor actions.

PubMed

Fee, Michale S

2012-01-01

In its simplest formulation, reinforcement learning is based on the idea that if an action taken in a particular context is followed by a favorable outcome, then, in the same context, the tendency to produce that action should be strengthened, or reinforced. While reinforcement learning forms the basis of many current theories of basal ganglia (BG) function, these models do not incorporate distinct computational roles for signals that convey context, and those that convey what action an animal takes. Recent experiments in the songbird suggest that vocal-related BG circuitry receives two functionally distinct excitatory inputs. One input is from a cortical region that carries context information about the current "time" in the motor sequence. The other is an efference copy of motor commands from a separate cortical brain region that generates vocal variability during learning. Based on these findings, I propose here a general model of vertebrate BG function that combines context information with a distinct motor efference copy signal. The signals are integrated by a learning rule in which efference copy inputs gate the potentiation of context inputs (but not efference copy inputs) onto medium spiny neurons in response to a rewarded action. The hypothesis is described in terms of a circuit that implements the learning of visually guided saccades. The model makes testable predictions about the anatomical and functional properties of hypothesized context and efference copy inputs to the striatum from both thalamic and cortical sources.

Microstimulation of the human substantia nigra alters reinforcement learning.

PubMed

Ramayya, Ashwin G; Misra, Amrit; Baltuch, Gordon H; Kahana, Michael J

2014-05-14

Animal studies have shown that substantia nigra (SN) dopaminergic (DA) neurons strengthen action-reward associations during reinforcement learning, but their role in human learning is not known. Here, we applied microstimulation in the SN of 11 patients undergoing deep brain stimulation surgery for the treatment of Parkinson's disease as they performed a two-alternative probability learning task in which rewards were contingent on stimuli, rather than actions. Subjects demonstrated decreased learning from reward trials that were accompanied by phasic SN microstimulation compared with reward trials without stimulation. Subjects who showed large decreases in learning also showed an increased bias toward repeating actions after stimulation trials; therefore, stimulation may have decreased learning by strengthening action-reward associations rather than stimulus-reward associations. Our findings build on previous studies implicating SN DA neurons in preferentially strengthening action-reward associations during reinforcement learning. Copyright © 2014 the authors 0270-6474/14/346887-09$15.00/0.

Genetic reinforcement learning through symbiotic evolution for fuzzy controller design.

PubMed

Juang, C F; Lin, J Y; Lin, C T

2000-01-01

An efficient genetic reinforcement learning algorithm for designing fuzzy controllers is proposed in this paper. The genetic algorithm (GA) adopted in this paper is based upon symbiotic evolution which, when applied to fuzzy controller design, complements the local mapping property of a fuzzy rule. Using this Symbiotic-Evolution-based Fuzzy Controller (SEFC) design method, the number of control trials, as well as consumed CPU time, are considerably reduced when compared to traditional GA-based fuzzy controller design methods and other types of genetic reinforcement learning schemes. Moreover, unlike traditional fuzzy controllers, which partition the input space into a grid, SEFC partitions the input space in a flexible way, thus creating fewer fuzzy rules. In SEFC, different types of fuzzy rules whose consequent parts are singletons, fuzzy sets, or linear equations (TSK-type fuzzy rules) are allowed. Further, the free parameters (e.g., centers and widths of membership functions) and fuzzy rules are all tuned automatically. For the TSK-type fuzzy rule especially, which put the proposed learning algorithm in use, only the significant input variables are selected to participate in the consequent of a rule. The proposed SEFC design method has been applied to different simulated control problems, including the cart-pole balancing system, a magnetic levitation system, and a water bath temperature control system. The proposed SEFC has been verified to be efficient and superior from these control problems, and from comparisons with some traditional GA-based fuzzy systems.

Use of Inverse Reinforcement Learning for Identity Prediction

NASA Technical Reports Server (NTRS)

Hayes, Roy; Bao, Jonathan; Beling, Peter; Horowitz, Barry

2011-01-01

We adopt Markov Decision Processes (MDP) to model sequential decision problems, which have the characteristic that the current decision made by a human decision maker has an uncertain impact on future opportunity. We hypothesize that the individuality of decision makers can be modeled as differences in the reward function under a common MDP model. A machine learning technique, Inverse Reinforcement Learning (IRL), was used to learn an individual's reward function based on limited observation of his or her decision choices. This work serves as an initial investigation for using IRL to analyze decision making, conducted through a human experiment in a cyber shopping environment. Specifically, the ability to determine the demographic identity of users is conducted through prediction analysis and supervised learning. The results show that IRL can be used to correctly identify participants, at a rate of 68% for gender and 66% for one of three college major categories.

Accelerating Multiagent Reinforcement Learning by Equilibrium Transfer.

PubMed

Hu, Yujing; Gao, Yang; An, Bo

2015-07-01

An important approach in multiagent reinforcement learning (MARL) is equilibrium-based MARL, which adopts equilibrium solution concepts in game theory and requires agents to play equilibrium strategies at each state. However, most existing equilibrium-based MARL algorithms cannot scale due to a large number of computationally expensive equilibrium computations (e.g., computing Nash equilibria is PPAD-hard) during learning. For the first time, this paper finds that during the learning process of equilibrium-based MARL, the one-shot games corresponding to each state's successive visits often have the same or similar equilibria (for some states more than 90% of games corresponding to successive visits have similar equilibria). Inspired by this observation, this paper proposes to use equilibrium transfer to accelerate equilibrium-based MARL. The key idea of equilibrium transfer is to reuse previously computed equilibria when each agent has a small incentive to deviate. By introducing transfer loss and transfer condition, a novel framework called equilibrium transfer-based MARL is proposed. We prove that although equilibrium transfer brings transfer loss, equilibrium-based MARL algorithms can still converge to an equilibrium policy under certain assumptions. Experimental results in widely used benchmarks (e.g., grid world game, soccer game, and wall game) show that the proposed framework: 1) not only significantly accelerates equilibrium-based MARL (up to 96.7% reduction in learning time), but also achieves higher average rewards than algorithms without equilibrium transfer and 2) scales significantly better than algorithms without equilibrium transfer when the state/action space grows and the number of agents increases.

Mastery Learning through Individualized Instruction: A Reinforcement Strategy

ERIC Educational Resources Information Center

Sagy, John; Ravi, R.; Ananthasayanam, R.

2009-01-01

The present study attempts to gauge the effect of individualized instructional methods as a reinforcement strategy for mastery learning. Among various individualized instructional methods, the study focuses on PIM (Programmed Instructional Method) and CAIM (Computer Assisted Instruction Method). Mastery learning is a process where students achieve…

'Proactive' use of cue-context congruence for building reinforcement learning's reward function.

PubMed

Zsuga, Judit; Biro, Klara; Tajti, Gabor; Szilasi, Magdolna Emma; Papp, Csaba; Juhasz, Bela; Gesztelyi, Rudolf

2016-10-28

Reinforcement learning is a fundamental form of learning that may be formalized using the Bellman equation. Accordingly an agent determines the state value as the sum of immediate reward and of the discounted value of future states. Thus the value of state is determined by agent related attributes (action set, policy, discount factor) and the agent's knowledge of the environment embodied by the reward function and hidden environmental factors given by the transition probability. The central objective of reinforcement learning is to solve these two functions outside the agent's control either using, or not using a model. In the present paper, using the proactive model of reinforcement learning we offer insight on how the brain creates simplified representations of the environment, and how these representations are organized to support the identification of relevant stimuli and action. Furthermore, we identify neurobiological correlates of our model by suggesting that the reward and policy functions, attributes of the Bellman equitation, are built by the orbitofrontal cortex (OFC) and the anterior cingulate cortex (ACC), respectively. Based on this we propose that the OFC assesses cue-context congruence to activate the most context frame. Furthermore given the bidirectional neuroanatomical link between the OFC and model-free structures, we suggest that model-based input is incorporated into the reward prediction error (RPE) signal, and conversely RPE signal may be used to update the reward-related information of context frames and the policy underlying action selection in the OFC and ACC, respectively. Furthermore clinical implications for cognitive behavioral interventions are discussed.

Attentional Selection Can Be Predicted by Reinforcement Learning of Task-relevant Stimulus Features Weighted by Value-independent Stickiness.

PubMed

Balcarras, Matthew; Ardid, Salva; Kaping, Daniel; Everling, Stefan; Womelsdorf, Thilo

2016-02-01

Attention includes processes that evaluate stimuli relevance, select the most relevant stimulus against less relevant stimuli, and bias choice behavior toward the selected information. It is not clear how these processes interact. Here, we captured these processes in a reinforcement learning framework applied to a feature-based attention task that required macaques to learn and update the value of stimulus features while ignoring nonrelevant sensory features, locations, and action plans. We found that value-based reinforcement learning mechanisms could account for feature-based attentional selection and choice behavior but required a value-independent stickiness selection process to explain selection errors while at asymptotic behavior. By comparing different reinforcement learning schemes, we found that trial-by-trial selections were best predicted by a model that only represents expected values for the task-relevant feature dimension, with nonrelevant stimulus features and action plans having only a marginal influence on covert selections. These findings show that attentional control subprocesses can be described by (1) the reinforcement learning of feature values within a restricted feature space that excludes irrelevant feature dimensions, (2) a stochastic selection process on feature-specific value representations, and (3) value-independent stickiness toward previous feature selections akin to perseveration in the motor domain. We speculate that these three mechanisms are implemented by distinct but interacting brain circuits and that the proposed formal account of feature-based stimulus selection will be important to understand how attentional subprocesses are implemented in primate brain networks.

Reinforcement Learning Explains Conditional Cooperation and Its Moody Cousin.

PubMed

Ezaki, Takahiro; Horita, Yutaka; Takezawa, Masanori; Masuda, Naoki

2016-07-01

Direct reciprocity, or repeated interaction, is a main mechanism to sustain cooperation under social dilemmas involving two individuals. For larger groups and networks, which are probably more relevant to understanding and engineering our society, experiments employing repeated multiplayer social dilemma games have suggested that humans often show conditional cooperation behavior and its moody variant. Mechanisms underlying these behaviors largely remain unclear. Here we provide a proximate account for this behavior by showing that individuals adopting a type of reinforcement learning, called aspiration learning, phenomenologically behave as conditional cooperator. By definition, individuals are satisfied if and only if the obtained payoff is larger than a fixed aspiration level. They reinforce actions that have resulted in satisfactory outcomes and anti-reinforce those yielding unsatisfactory outcomes. The results obtained in the present study are general in that they explain extant experimental results obtained for both so-called moody and non-moody conditional cooperation, prisoner's dilemma and public goods games, and well-mixed groups and networks. Different from the previous theory, individuals are assumed to have no access to information about what other individuals are doing such that they cannot explicitly use conditional cooperation rules. In this sense, myopic aspiration learning in which the unconditional propensity of cooperation is modulated in every discrete time step explains conditional behavior of humans. Aspiration learners showing (moody) conditional cooperation obeyed a noisy GRIM-like strategy. This is different from the Pavlov, a reinforcement learning strategy promoting mutual cooperation in two-player situations.

Adolescent-specific patterns of behavior and neural activity during social reinforcement learning

PubMed Central

Jones, Rebecca M.; Somerville, Leah H.; Li, Jian; Ruberry, Erika J.; Powers, Alisa; Mehta, Natasha; Dyke, Jonathan; Casey, BJ

2014-01-01

Humans are sophisticated social beings. Social cues from others are exceptionally salient, particularly during adolescence. Understanding how adolescents interpret and learn from variable social signals can provide insight into the observed shift in social sensitivity during this period. The current study tested 120 participants between the ages of 8 and 25 years on a social reinforcement learning task where the probability of receiving positive social feedback was parametrically manipulated. Seventy-eight of these participants completed the task during fMRI scanning. Modeling trial-by-trial learning, children and adults showed higher positive learning rates than adolescents, suggesting that adolescents demonstrated less differentiation in their reaction times for peers who provided more positive feedback. Forming expectations about receiving positive social reinforcement correlated with neural activity within the medial prefrontal cortex and ventral striatum across age. Adolescents, unlike children and adults, showed greater insular activity during positive prediction error learning and increased activity in the supplementary motor cortex and the putamen when receiving positive social feedback regardless of the expected outcome, suggesting that peer approval may motivate adolescents towards action. While different amounts of positive social reinforcement enhanced learning in children and adults, all positive social reinforcement equally motivated adolescents. Together, these findings indicate that sensitivity to peer approval during adolescence goes beyond simple reinforcement theory accounts and suggests possible explanations for how peers may motivate adolescent behavior. PMID:24550063

Adolescent-specific patterns of behavior and neural activity during social reinforcement learning.

PubMed

Jones, Rebecca M; Somerville, Leah H; Li, Jian; Ruberry, Erika J; Powers, Alisa; Mehta, Natasha; Dyke, Jonathan; Casey, B J

2014-06-01

Humans are sophisticated social beings. Social cues from others are exceptionally salient, particularly during adolescence. Understanding how adolescents interpret and learn from variable social signals can provide insight into the observed shift in social sensitivity during this period. The present study tested 120 participants between the ages of 8 and 25 years on a social reinforcement learning task where the probability of receiving positive social feedback was parametrically manipulated. Seventy-eight of these participants completed the task during fMRI scanning. Modeling trial-by-trial learning, children and adults showed higher positive learning rates than did adolescents, suggesting that adolescents demonstrated less differentiation in their reaction times for peers who provided more positive feedback. Forming expectations about receiving positive social reinforcement correlated with neural activity within the medial prefrontal cortex and ventral striatum across age. Adolescents, unlike children and adults, showed greater insular activity during positive prediction error learning and increased activity in the supplementary motor cortex and the putamen when receiving positive social feedback regardless of the expected outcome, suggesting that peer approval may motivate adolescents toward action. While different amounts of positive social reinforcement enhanced learning in children and adults, all positive social reinforcement equally motivated adolescents. Together, these findings indicate that sensitivity to peer approval during adolescence goes beyond simple reinforcement theory accounts and suggest possible explanations for how peers may motivate adolescent behavior.

The Computational Development of Reinforcement Learning during Adolescence

PubMed Central

Palminteri, Stefano; Coricelli, Giorgio; Blakemore, Sarah-Jayne

2016-01-01

Adolescence is a period of life characterised by changes in learning and decision-making. Learning and decision-making do not rely on a unitary system, but instead require the coordination of different cognitive processes that can be mathematically formalised as dissociable computational modules. Here, we aimed to trace the developmental time-course of the computational modules responsible for learning from reward or punishment, and learning from counterfactual feedback. Adolescents and adults carried out a novel reinforcement learning paradigm in which participants learned the association between cues and probabilistic outcomes, where the outcomes differed in valence (reward versus punishment) and feedback was either partial or complete (either the outcome of the chosen option only, or the outcomes of both the chosen and unchosen option, were displayed). Computational strategies changed during development: whereas adolescents’ behaviour was better explained by a basic reinforcement learning algorithm, adults’ behaviour integrated increasingly complex computational features, namely a counterfactual learning module (enabling enhanced performance in the presence of complete feedback) and a value contextualisation module (enabling symmetrical reward and punishment learning). Unlike adults, adolescent performance did not benefit from counterfactual (complete) feedback. In addition, while adults learned symmetrically from both reward and punishment, adolescents learned from reward but were less likely to learn from punishment. This tendency to rely on rewards and not to consider alternative consequences of actions might contribute to our understanding of decision-making in adolescence. PMID:27322574

Reinforcement learning agents providing advice in complex video games

NASA Astrophysics Data System (ADS)

Taylor, Matthew E.; Carboni, Nicholas; Fachantidis, Anestis; Vlahavas, Ioannis; Torrey, Lisa

2014-01-01

This article introduces a teacher-student framework for reinforcement learning, synthesising and extending material that appeared in conference proceedings [Torrey, L., & Taylor, M. E. (2013)]. Teaching on a budget: Agents advising agents in reinforcement learning. {Proceedings of the international conference on autonomous agents and multiagent systems}] and in a non-archival workshop paper [Carboni, N., &Taylor, M. E. (2013, May)]. Preliminary results for 1 vs. 1 tactics in StarCraft. {Proceedings of the adaptive and learning agents workshop (at AAMAS-13)}]. In this framework, a teacher agent instructs a student agent by suggesting actions the student should take as it learns. However, the teacher may only give such advice a limited number of times. We present several novel algorithms that teachers can use to budget their advice effectively, and we evaluate them in two complex video games: StarCraft and Pac-Man. Our results show that the same amount of advice, given at different moments, can have different effects on student learning, and that teachers can significantly affect student learning even when students use different learning methods and state representations.

Applications of Deep Learning and Reinforcement Learning to Biological Data.

PubMed

Mahmud, Mufti; Kaiser, Mohammed Shamim; Hussain, Amir; Vassanelli, Stefano

2018-06-01

Rapid advances in hardware-based technologies during the past decades have opened up new possibilities for life scientists to gather multimodal data in various application domains, such as omics, bioimaging, medical imaging, and (brain/body)-machine interfaces. These have generated novel opportunities for development of dedicated data-intensive machine learning techniques. In particular, recent research in deep learning (DL), reinforcement learning (RL), and their combination (deep RL) promise to revolutionize the future of artificial intelligence. The growth in computational power accompanied by faster and increased data storage, and declining computing costs have already allowed scientists in various fields to apply these techniques on data sets that were previously intractable owing to their size and complexity. This paper provides a comprehensive survey on the application of DL, RL, and deep RL techniques in mining biological data. In addition, we compare the performances of DL techniques when applied to different data sets across various application domains. Finally, we outline open issues in this challenging research area and discuss future development perspectives.

Cocaine addiction as a homeostatic reinforcement learning disorder.

PubMed

Keramati, Mehdi; Durand, Audrey; Girardeau, Paul; Gutkin, Boris; Ahmed, Serge H

2017-03-01

Drug addiction implicates both reward learning and homeostatic regulation mechanisms of the brain. This has stimulated 2 partially successful theoretical perspectives on addiction. Many important aspects of addiction, however, remain to be explained within a single, unified framework that integrates the 2 mechanisms. Building upon a recently developed homeostatic reinforcement learning theory, the authors focus on a key transition stage of addiction that is well modeled in animals, escalation of drug use, and propose a computational theory of cocaine addiction where cocaine reinforces behavior due to its rapid homeostatic corrective effect, whereas its chronic use induces slow and long-lasting changes in homeostatic setpoint. Simulations show that our new theory accounts for key behavioral and neurobiological features of addiction, most notably, escalation of cocaine use, drug-primed craving and relapse, individual differences underlying dose-response curves, and dopamine D2-receptor downregulation in addicts. The theory also generates unique predictions about cocaine self-administration behavior in rats that are confirmed by new experimental results. Viewing addiction as a homeostatic reinforcement learning disorder coherently explains many behavioral and neurobiological aspects of the transition to cocaine addiction, and suggests a new perspective toward understanding addiction. (PsycINFO Database Record (c) 2017 APA, all rights reserved).

Time-Extended Policies in Mult-Agent Reinforcement Learning

NASA Technical Reports Server (NTRS)

Tumer, Kagan; Agogino, Adrian K.

2004-01-01

Reinforcement learning methods perform well in many domains where a single agent needs to take a sequence of actions to perform a task. These methods use sequences of single-time-step rewards to create a policy that tries to maximize a time-extended utility, which is a (possibly discounted) sum of these rewards. In this paper we build on our previous work showing how these methods can be extended to a multi-agent environment where each agent creates its own policy that works towards maximizing a time-extended global utility over all agents actions. We show improved methods for creating time-extended utilities for the agents that are both "aligned" with the global utility and "learnable." We then show how to crate single-time-step rewards while avoiding the pi fall of having rewards aligned with the global reward leading to utilities not aligned with the global utility. Finally, we apply these reward functions to the multi-agent Gridworld problem. We explicitly quantify a utility's learnability and alignment, and show that reinforcement learning agents using the prescribed reward functions successfully tradeoff learnability and alignment. As a result they outperform both global (e.g., team games ) and local (e.g., "perfectly learnable" ) reinforcement learning solutions by as much as an order of magnitude.

Scheduled power tracking control of the wind-storage hybrid system based on the reinforcement learning theory

NASA Astrophysics Data System (ADS)

Li, Ze

2017-09-01

In allusion to the intermittency and uncertainty of the wind electricity, energy storage and wind generator are combined into a hybrid system to improve the controllability of the output power. A scheduled power tracking control method is proposed based on the reinforcement learning theory and Q-learning algorithm. In this method, the state space of the environment is formed with two key factors, i.e. the state of charge of the energy storage and the difference value between the actual wind power and scheduled power, the feasible action is the output power of the energy storage, and the corresponding immediate rewarding function is designed to reflect the rationality of the control action. By interacting with the environment and learning from the immediate reward, the optimal control strategy is gradually formed. After that, it could be applied to the scheduled power tracking control of the hybrid system. Finally, the rationality and validity of the method are verified through simulation examples.

What is the optimal task difficulty for reinforcement learning of brain self-regulation?

PubMed

Bauer, Robert; Vukelić, Mathias; Gharabaghi, Alireza

2016-09-01

The balance between action and reward during neurofeedback may influence reinforcement learning of brain self-regulation. Eleven healthy volunteers participated in three runs of motor imagery-based brain-machine interface feedback where a robot passively opened the hand contingent to β-band modulation. For each run, the β-desynchronization threshold to initiate the hand robot movement increased in difficulty (low, moderate, and demanding). In this context, the incentive to learn was estimated by the change of reward per action, operationalized as the change in reward duration per movement onset. Variance analysis revealed a significant interaction between threshold difficulty and the relationship between reward duration and number of movement onsets (p<0.001), indicating a negative learning incentive for low difficulty, but a positive learning incentive for moderate and demanding runs. Exploration of different thresholds in the same data set indicated that the learning incentive peaked at higher thresholds than the threshold which resulted in maximum classification accuracy. Specificity is more important than sensitivity of neurofeedback for reinforcement learning of brain self-regulation. Learning efficiency requires adequate challenge by neurofeedback interventions. Copyright © 2016 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.

Place preference and vocal learning rely on distinct reinforcers in songbirds.

PubMed

Murdoch, Don; Chen, Ruidong; Goldberg, Jesse H

2018-04-30

In reinforcement learning (RL) agents are typically tasked with maximizing a single objective function such as reward. But it remains poorly understood how agents might pursue distinct objectives at once. In machines, multiobjective RL can be achieved by dividing a single agent into multiple sub-agents, each of which is shaped by agent-specific reinforcement, but it remains unknown if animals adopt this strategy. Here we use songbirds to test if navigation and singing, two behaviors with distinct objectives, can be differentially reinforced. We demonstrate that strobe flashes aversively condition place preference but not song syllables. Brief noise bursts aversively condition song syllables but positively reinforce place preference. Thus distinct behavior-generating systems, or agencies, within a single animal can be shaped by correspondingly distinct reinforcement signals. Our findings suggest that spatially segregated vocal circuits can solve a credit assignment problem associated with multiobjective learning.

Maximization of Learning Speed Due to Neuronal Redundancy in Reinforcement Learning

NASA Astrophysics Data System (ADS)

Takiyama, Ken

2016-11-01

Adaptable neural activity contributes to the flexibility of human behavior, which is optimized in situations such as motor learning and decision making. Although learning signals in motor learning and decision making are low-dimensional, neural activity, which is very high dimensional, must be modified to achieve optimal performance based on the low-dimensional signal, resulting in a severe credit-assignment problem. Despite this problem, the human brain contains a vast number of neurons, leaving an open question: what is the functional significance of the huge number of neurons? Here, I address this question by analyzing a redundant neural network with a reinforcement-learning algorithm in which the numbers of neurons and output units are N and M, respectively. Because many combinations of neural activity can generate the same output under the condition of N ≫ M, I refer to the index N - M as neuronal redundancy. Although greater neuronal redundancy makes the credit-assignment problem more severe, I demonstrate that a greater degree of neuronal redundancy facilitates learning speed. Thus, in an apparent contradiction of the credit-assignment problem, I propose the hypothesis that a functional role of a huge number of neurons or a huge degree of neuronal redundancy is to facilitate learning speed.

Pleasurable music affects reinforcement learning according to the listener

PubMed Central

Gold, Benjamin P.; Frank, Michael J.; Bogert, Brigitte; Brattico, Elvira

2013-01-01

Mounting evidence links the enjoyment of music to brain areas implicated in emotion and the dopaminergic reward system. In particular, dopamine release in the ventral striatum seems to play a major role in the rewarding aspect of music listening. Striatal dopamine also influences reinforcement learning, such that subjects with greater dopamine efficacy learn better to approach rewards while those with lesser dopamine efficacy learn better to avoid punishments. In this study, we explored the practical implications of musical pleasure through its ability to facilitate reinforcement learning via non-pharmacological dopamine elicitation. Subjects from a wide variety of musical backgrounds chose a pleasurable and a neutral piece of music from an experimenter-compiled database, and then listened to one or both of these pieces (according to pseudo-random group assignment) as they performed a reinforcement learning task dependent on dopamine transmission. We assessed musical backgrounds as well as typical listening patterns with the new Helsinki Inventory of Music and Affective Behaviors (HIMAB), and separately investigated behavior for the training and test phases of the learning task. Subjects with more musical experience trained better with neutral music and tested better with pleasurable music, while those with less musical experience exhibited the opposite effect. HIMAB results regarding listening behaviors and subjective music ratings indicate that these effects arose from different listening styles: namely, more affective listening in non-musicians and more analytical listening in musicians. In conclusion, musical pleasure was able to influence task performance, and the shape of this effect depended on group and individual factors. These findings have implications in affective neuroscience, neuroaesthetics, learning, and music therapy. PMID:23970875

Reinforcement learning in depression: A review of computational research.

PubMed

Chen, Chong; Takahashi, Taiki; Nakagawa, Shin; Inoue, Takeshi; Kusumi, Ichiro

2015-08-01

Despite being considered primarily a mood disorder, major depressive disorder (MDD) is characterized by cognitive and decision making deficits. Recent research has employed computational models of reinforcement learning (RL) to address these deficits. The computational approach has the advantage in making explicit predictions about learning and behavior, specifying the process parameters of RL, differentiating between model-free and model-based RL, and the computational model-based functional magnetic resonance imaging and electroencephalography. With these merits there has been an emerging field of computational psychiatry and here we review specific studies that focused on MDD. Considerable evidence suggests that MDD is associated with impaired brain signals of reward prediction error and expected value ('wanting'), decreased reward sensitivity ('liking') and/or learning (be it model-free or model-based), etc., although the causality remains unclear. These parameters may serve as valuable intermediate phenotypes of MDD, linking general clinical symptoms to underlying molecular dysfunctions. We believe future computational research at clinical, systems, and cellular/molecular/genetic levels will propel us toward a better understanding of the disease. Copyright © 2015 Elsevier Ltd. All rights reserved.

Towards autonomous neuroprosthetic control using Hebbian reinforcement learning.

PubMed

Mahmoudi, Babak; Pohlmeyer, Eric A; Prins, Noeline W; Geng, Shijia; Sanchez, Justin C

2013-12-01

Our goal was to design an adaptive neuroprosthetic controller that could learn the mapping from neural states to prosthetic actions and automatically adjust adaptation using only a binary evaluative feedback as a measure of desirability/undesirability of performance. Hebbian reinforcement learning (HRL) in a connectionist network was used for the design of the adaptive controller. The method combines the efficiency of supervised learning with the generality of reinforcement learning. The convergence properties of this approach were studied using both closed-loop control simulations and open-loop simulations that used primate neural data from robot-assisted reaching tasks. The HRL controller was able to perform classification and regression tasks using its episodic and sequential learning modes, respectively. In our experiments, the HRL controller quickly achieved convergence to an effective control policy, followed by robust performance. The controller also automatically stopped adapting the parameters after converging to a satisfactory control policy. Additionally, when the input neural vector was reorganized, the controller resumed adaptation to maintain performance. By estimating an evaluative feedback directly from the user, the HRL control algorithm may provide an efficient method for autonomous adaptation of neuroprosthetic systems. This method may enable the user to teach the controller the desired behavior using only a simple feedback signal.

Neurofeedback in Learning Disabled Children: Visual versus Auditory Reinforcement.

PubMed

Fernández, Thalía; Bosch-Bayard, Jorge; Harmony, Thalía; Caballero, María I; Díaz-Comas, Lourdes; Galán, Lídice; Ricardo-Garcell, Josefina; Aubert, Eduardo; Otero-Ojeda, Gloria

2016-03-01

Children with learning disabilities (LD) frequently have an EEG characterized by an excess of theta and a deficit of alpha activities. NFB using an auditory stimulus as reinforcer has proven to be a useful tool to treat LD children by positively reinforcing decreases of the theta/alpha ratio. The aim of the present study was to optimize the NFB procedure by comparing the efficacy of visual (with eyes open) versus auditory (with eyes closed) reinforcers. Twenty LD children with an abnormally high theta/alpha ratio were randomly assigned to the Auditory or the Visual group, where a 500 Hz tone or a visual stimulus (a white square), respectively, was used as a positive reinforcer when the value of the theta/alpha ratio was reduced. Both groups had signs consistent with EEG maturation, but only the Auditory Group showed behavioral/cognitive improvements. In conclusion, the auditory reinforcer was more efficacious in reducing the theta/alpha ratio, and it improved the cognitive abilities more than the visual reinforcer.

A parameter control method in reinforcement learning to rapidly follow unexpected environmental changes.

PubMed

Murakoshi, Kazushi; Mizuno, Junya

2004-11-01

In order to rapidly follow unexpected environmental changes, we propose a parameter control method in reinforcement learning that changes each of learning parameters in appropriate directions. We determine each appropriate direction on the basis of relationships between behaviors and neuromodulators by considering an emergency as a key word. Computer experiments show that the agents using our proposed method could rapidly respond to unexpected environmental changes, not depending on either two reinforcement learning algorithms (Q-learning and actor-critic (AC) architecture) or two learning problems (discontinuous and continuous state-action problems).

Action-Driven Visual Object Tracking With Deep Reinforcement Learning.

PubMed

Yun, Sangdoo; Choi, Jongwon; Yoo, Youngjoon; Yun, Kimin; Choi, Jin Young

2018-06-01

In this paper, we propose an efficient visual tracker, which directly captures a bounding box containing the target object in a video by means of sequential actions learned using deep neural networks. The proposed deep neural network to control tracking actions is pretrained using various training video sequences and fine-tuned during actual tracking for online adaptation to a change of target and background. The pretraining is done by utilizing deep reinforcement learning (RL) as well as supervised learning. The use of RL enables even partially labeled data to be successfully utilized for semisupervised learning. Through the evaluation of the object tracking benchmark data set, the proposed tracker is validated to achieve a competitive performance at three times the speed of existing deep network-based trackers. The fast version of the proposed method, which operates in real time on graphics processing unit, outperforms the state-of-the-art real-time trackers with an accuracy improvement of more than 8%.

Reinforcement Learning with Autonomous Small Unmanned Aerial Vehicles in Cluttered Environments

NASA Technical Reports Server (NTRS)

Tran, Loc; Cross, Charles; Montague, Gilbert; Motter, Mark; Neilan, James; Qualls, Garry; Rothhaar, Paul; Trujillo, Anna; Allen, B. Danette

2015-01-01

We present ongoing work in the Autonomy Incubator at NASA Langley Research Center (LaRC) exploring the efficacy of a data set aggregation approach to reinforcement learning for small unmanned aerial vehicle (sUAV) flight in dense and cluttered environments with reactive obstacle avoidance. The goal is to learn an autonomous flight model using training experiences from a human piloting a sUAV around static obstacles. The training approach uses video data from a forward-facing camera that records the human pilot's flight. Various computer vision based features are extracted from the video relating to edge and gradient information. The recorded human-controlled inputs are used to train an autonomous control model that correlates the extracted feature vector to a yaw command. As part of the reinforcement learning approach, the autonomous control model is iteratively updated with feedback from a human agent who corrects undesired model output. This data driven approach to autonomous obstacle avoidance is explored for simulated forest environments furthering autonomous flight under the tree canopy research. This enables flight in previously inaccessible environments which are of interest to NASA researchers in Earth and Atmospheric sciences.

An Energy-Efficient Spectrum-Aware Reinforcement Learning-Based Clustering Algorithm for Cognitive Radio Sensor Networks

PubMed Central

Mustapha, Ibrahim; Ali, Borhanuddin Mohd; Rasid, Mohd Fadlee A.; Sali, Aduwati; Mohamad, Hafizal

2015-01-01

It is well-known that clustering partitions network into logical groups of nodes in order to achieve energy efficiency and to enhance dynamic channel access in cognitive radio through cooperative sensing. While the topic of energy efficiency has been well investigated in conventional wireless sensor networks, the latter has not been extensively explored. In this paper, we propose a reinforcement learning-based spectrum-aware clustering algorithm that allows a member node to learn the energy and cooperative sensing costs for neighboring clusters to achieve an optimal solution. Each member node selects an optimal cluster that satisfies pairwise constraints, minimizes network energy consumption and enhances channel sensing performance through an exploration technique. We first model the network energy consumption and then determine the optimal number of clusters for the network. The problem of selecting an optimal cluster is formulated as a Markov Decision Process (MDP) in the algorithm and the obtained simulation results show convergence, learning and adaptability of the algorithm to dynamic environment towards achieving an optimal solution. Performance comparisons of our algorithm with the Groupwise Spectrum Aware (GWSA)-based algorithm in terms of Sum of Square Error (SSE), complexity, network energy consumption and probability of detection indicate improved performance from the proposed approach. The results further reveal that an energy savings of 9% and a significant Primary User (PU) detection improvement can be achieved with the proposed approach. PMID:26287191

An Energy-Efficient Spectrum-Aware Reinforcement Learning-Based Clustering Algorithm for Cognitive Radio Sensor Networks.

PubMed

Mustapha, Ibrahim; Mohd Ali, Borhanuddin; Rasid, Mohd Fadlee A; Sali, Aduwati; Mohamad, Hafizal

2015-08-13

It is well-known that clustering partitions network into logical groups of nodes in order to achieve energy efficiency and to enhance dynamic channel access in cognitive radio through cooperative sensing. While the topic of energy efficiency has been well investigated in conventional wireless sensor networks, the latter has not been extensively explored. In this paper, we propose a reinforcement learning-based spectrum-aware clustering algorithm that allows a member node to learn the energy and cooperative sensing costs for neighboring clusters to achieve an optimal solution. Each member node selects an optimal cluster that satisfies pairwise constraints, minimizes network energy consumption and enhances channel sensing performance through an exploration technique. We first model the network energy consumption and then determine the optimal number of clusters for the network. The problem of selecting an optimal cluster is formulated as a Markov Decision Process (MDP) in the algorithm and the obtained simulation results show convergence, learning and adaptability of the algorithm to dynamic environment towards achieving an optimal solution. Performance comparisons of our algorithm with the Groupwise Spectrum Aware (GWSA)-based algorithm in terms of Sum of Square Error (SSE), complexity, network energy consumption and probability of detection indicate improved performance from the proposed approach. The results further reveal that an energy savings of 9% and a significant Primary User (PU) detection improvement can be achieved with the proposed approach.

Cerebellar and Prefrontal Cortex Contributions to Adaptation, Strategies, and Reinforcement Learning

PubMed Central

Taylor, Jordan A.; Ivry, Richard B.

2014-01-01

Traditionally, motor learning has been studied as an implicit learning process, one in which movement errors are used to improve performance in a continuous, gradual manner. The cerebellum figures prominently in this literature given well-established ideas about the role of this system in error-based learning and the production of automatized skills. Recent developments have brought into focus the relevance of multiple learning mechanisms for sensorimotor learning. These include processes involving repetition, reinforcement learning, and strategy utilization. We examine these developments, considering their implications for understanding cerebellar function and how this structure interacts with other neural systems to support motor learning. Converging lines of evidence from behavioral, computational, and neuropsychological studies suggest a fundamental distinction between processes that use error information to improve action execution or action selection. While the cerebellum is clearly linked to the former, its role in the latter remains an open question. PMID:24916295

The combination of appetitive and aversive reinforcers and the nature of their interaction during auditory learning.

PubMed

Ilango, A; Wetzel, W; Scheich, H; Ohl, F W

2010-03-31

Learned changes in behavior can be elicited by either appetitive or aversive reinforcers. It is, however, not clear whether the two types of motivation, (approaching appetitive stimuli and avoiding aversive stimuli) drive learning in the same or different ways, nor is their interaction understood in situations where the two types are combined in a single experiment. To investigate this question we have developed a novel learning paradigm for Mongolian gerbils, which not only allows rewards and punishments to be presented in isolation or in combination with each other, but also can use these opposite reinforcers to drive the same learned behavior. Specifically, we studied learning of tone-conditioned hurdle crossing in a shuttle box driven by either an appetitive reinforcer (brain stimulation reward) or an aversive reinforcer (electrical footshock), or by a combination of both. Combination of the two reinforcers potentiated speed of acquisition, led to maximum possible performance, and delayed extinction as compared to either reinforcer alone. Additional experiments, using partial reinforcement protocols and experiments in which one of the reinforcers was omitted after the animals had been previously trained with the combination of both reinforcers, indicated that appetitive and aversive reinforcers operated together but acted in different ways: in this particular experimental context, punishment appeared to be more effective for initial acquisition and reward more effective to maintain a high level of conditioned responses (CRs). The results imply that learning mechanisms in problem solving were maximally effective when the initial punishment of mistakes was combined with the subsequent rewarding of correct performance. Copyright 2010 IBRO. Published by Elsevier Ltd. All rights reserved.

Reinforcement learning for resource allocation in LEO satellite networks.

PubMed

Usaha, Wipawee; Barria, Javier A

2007-06-01

In this paper, we develop and assess online decision-making algorithms for call admission and routing for low Earth orbit (LEO) satellite networks. It has been shown in a recent paper that, in a LEO satellite system, a semi-Markov decision process formulation of the call admission and routing problem can achieve better performance in terms of an average revenue function than existing routing methods. However, the conventional dynamic programming (DP) numerical solution becomes prohibited as the problem size increases. In this paper, two solution methods based on reinforcement learning (RL) are proposed in order to circumvent the computational burden of DP. The first method is based on an actor-critic method with temporal-difference (TD) learning. The second method is based on a critic-only method, called optimistic TD learning. The algorithms enhance performance in terms of requirements in storage, computational complexity and computational time, and in terms of an overall long-term average revenue function that penalizes blocked calls. Numerical studies are carried out, and the results obtained show that the RL framework can achieve up to 56% higher average revenue over existing routing methods used in LEO satellite networks with reasonable storage and computational requirements.

Leveraging Intermediaries to Expand Work-Based Learning. Connecting the Classroom to Careers: The State's Role in Work-Based Learning

ERIC Educational Resources Information Center

Advance CTE: State Leaders Connecting Learning to Work, 2016

2016-01-01

Work-based learning is an educational strategy that offers students an opportunity to reinforce and deepen their classroom learning, explore future career fields and demonstrate their skills in an authentic setting. Managing work-based learning requires layers of coordination, which is typically done by an individual or organizational…

Synchronization of Chaotic Systems without Direct Connections Using Reinforcement Learning

NASA Astrophysics Data System (ADS)

Sato, Norihisa; Adachi, Masaharu

In this paper, we propose a control method for the synchronization of chaotic systems that does not require the systems to be connected, unlike existing methods such as that proposed by Pecora and Carroll in 1990. The method is based on the reinforcement learning algorithm. We apply our method to two discrete-time chaotic systems with mismatched parameters and achieve M step delay synchronization. Moreover, we extend the proposed method to the synchronization of continuous-time chaotic systems.

Cerebellar and prefrontal cortex contributions to adaptation, strategies, and reinforcement learning.

PubMed

Taylor, Jordan A; Ivry, Richard B

2014-01-01

Traditionally, motor learning has been studied as an implicit learning process, one in which movement errors are used to improve performance in a continuous, gradual manner. The cerebellum figures prominently in this literature given well-established ideas about the role of this system in error-based learning and the production of automatized skills. Recent developments have brought into focus the relevance of multiple learning mechanisms for sensorimotor learning. These include processes involving repetition, reinforcement learning, and strategy utilization. We examine these developments, considering their implications for understanding cerebellar function and how this structure interacts with other neural systems to support motor learning. Converging lines of evidence from behavioral, computational, and neuropsychological studies suggest a fundamental distinction between processes that use error information to improve action execution or action selection. While the cerebellum is clearly linked to the former, its role in the latter remains an open question. © 2014 Elsevier B.V. All rights reserved.

Context transfer in reinforcement learning using action-value functions.

PubMed

Mousavi, Amin; Nadjar Araabi, Babak; Nili Ahmadabadi, Majid

2014-01-01

This paper discusses the notion of context transfer in reinforcement learning tasks. Context transfer, as defined in this paper, implies knowledge transfer between source and target tasks that share the same environment dynamics and reward function but have different states or action spaces. In other words, the agents learn the same task while using different sensors and actuators. This requires the existence of an underlying common Markov decision process (MDP) to which all the agents' MDPs can be mapped. This is formulated in terms of the notion of MDP homomorphism. The learning framework is Q-learning. To transfer the knowledge between these tasks, the feature space is used as a translator and is expressed as a partial mapping between the state-action spaces of different tasks. The Q-values learned during the learning process of the source tasks are mapped to the sets of Q-values for the target task. These transferred Q-values are merged together and used to initialize the learning process of the target task. An interval-based approach is used to represent and merge the knowledge of the source tasks. Empirical results show that the transferred initialization can be beneficial to the learning process of the target task.

Context Transfer in Reinforcement Learning Using Action-Value Functions

PubMed Central

Mousavi, Amin; Nadjar Araabi, Babak; Nili Ahmadabadi, Majid

2014-01-01

This paper discusses the notion of context transfer in reinforcement learning tasks. Context transfer, as defined in this paper, implies knowledge transfer between source and target tasks that share the same environment dynamics and reward function but have different states or action spaces. In other words, the agents learn the same task while using different sensors and actuators. This requires the existence of an underlying common Markov decision process (MDP) to which all the agents' MDPs can be mapped. This is formulated in terms of the notion of MDP homomorphism. The learning framework is Q-learning. To transfer the knowledge between these tasks, the feature space is used as a translator and is expressed as a partial mapping between the state-action spaces of different tasks. The Q-values learned during the learning process of the source tasks are mapped to the sets of Q-values for the target task. These transferred Q-values are merged together and used to initialize the learning process of the target task. An interval-based approach is used to represent and merge the knowledge of the source tasks. Empirical results show that the transferred initialization can be beneficial to the learning process of the target task. PMID:25610457

Application of fuzzy logic-neural network based reinforcement learning to proximity and docking operations

NASA Technical Reports Server (NTRS)

Jani, Yashvant

1992-01-01

As part of the Research Institute for Computing and Information Systems (RICIS) activity, the reinforcement learning techniques developed at Ames Research Center are being applied to proximity and docking operations using the Shuttle and Solar Max satellite simulation. This activity is carried out in the software technology laboratory utilizing the Orbital Operations Simulator (OOS). This interim report provides the status of the project and outlines the future plans.

A Robust Cooperated Control Method with Reinforcement Learning and Adaptive H∞ Control

NASA Astrophysics Data System (ADS)

Obayashi, Masanao; Uchiyama, Shogo; Kuremoto, Takashi; Kobayashi, Kunikazu

This study proposes a robust cooperated control method combining reinforcement learning with robust control to control the system. A remarkable characteristic of the reinforcement learning is that it doesn't require model formula, however, it doesn't guarantee the stability of the system. On the other hand, robust control system guarantees stability and robustness, however, it requires model formula. We employ both the actor-critic method which is a kind of reinforcement learning with minimal amount of computation to control continuous valued actions and the traditional robust control, that is, H∞ control. The proposed system was compared method with the conventional control method, that is, the actor-critic only used, through the computer simulation of controlling the angle and the position of a crane system, and the simulation result showed the effectiveness of the proposed method.

Spared internal but impaired external reward prediction error signals in major depressive disorder during reinforcement learning.

PubMed

Bakic, Jasmina; Pourtois, Gilles; Jepma, Marieke; Duprat, Romain; De Raedt, Rudi; Baeken, Chris

2017-01-01

Major depressive disorder (MDD) creates debilitating effects on a wide range of cognitive functions, including reinforcement learning (RL). In this study, we sought to assess whether reward processing as such, or alternatively the complex interplay between motivation and reward might potentially account for the abnormal reward-based learning in MDD. A total of 35 treatment resistant MDD patients and 44 age matched healthy controls (HCs) performed a standard probabilistic learning task. RL was titrated using behavioral, computational modeling and event-related brain potentials (ERPs) data. MDD patients showed comparable learning rate compared to HCs. However, they showed decreased lose-shift responses as well as blunted subjective evaluations of the reinforcers used during the task, relative to HCs. Moreover, MDD patients showed normal internal (at the level of error-related negativity, ERN) but abnormal external (at the level of feedback-related negativity, FRN) reward prediction error (RPE) signals during RL, selectively when additional efforts had to be made to establish learning. Collectively, these results lend support to the assumption that MDD does not impair reward processing per se during RL. Instead, it seems to alter the processing of the emotional value of (external) reinforcers during RL, when additional intrinsic motivational processes have to be engaged. © 2016 Wiley Periodicals, Inc.

Reinforcement Learning Using a Continuous Time Actor-Critic Framework with Spiking Neurons

PubMed Central

Frémaux, Nicolas; Sprekeler, Henning; Gerstner, Wulfram

2013-01-01

Animals repeat rewarded behaviors, but the physiological basis of reward-based learning has only been partially elucidated. On one hand, experimental evidence shows that the neuromodulator dopamine carries information about rewards and affects synaptic plasticity. On the other hand, the theory of reinforcement learning provides a framework for reward-based learning. Recent models of reward-modulated spike-timing-dependent plasticity have made first steps towards bridging the gap between the two approaches, but faced two problems. First, reinforcement learning is typically formulated in a discrete framework, ill-adapted to the description of natural situations. Second, biologically plausible models of reward-modulated spike-timing-dependent plasticity require precise calculation of the reward prediction error, yet it remains to be shown how this can be computed by neurons. Here we propose a solution to these problems by extending the continuous temporal difference (TD) learning of Doya (2000) to the case of spiking neurons in an actor-critic network operating in continuous time, and with continuous state and action representations. In our model, the critic learns to predict expected future rewards in real time. Its activity, together with actual rewards, conditions the delivery of a neuromodulatory TD signal to itself and to the actor, which is responsible for action choice. In simulations, we show that such an architecture can solve a Morris water-maze-like navigation task, in a number of trials consistent with reported animal performance. We also use our model to solve the acrobot and the cartpole problems, two complex motor control tasks. Our model provides a plausible way of computing reward prediction error in the brain. Moreover, the analytically derived learning rule is consistent with experimental evidence for dopamine-modulated spike-timing-dependent plasticity. PMID:23592970

Reinforcement learning using a continuous time actor-critic framework with spiking neurons.

PubMed

Frémaux, Nicolas; Sprekeler, Henning; Gerstner, Wulfram

2013-04-01

Animals repeat rewarded behaviors, but the physiological basis of reward-based learning has only been partially elucidated. On one hand, experimental evidence shows that the neuromodulator dopamine carries information about rewards and affects synaptic plasticity. On the other hand, the theory of reinforcement learning provides a framework for reward-based learning. Recent models of reward-modulated spike-timing-dependent plasticity have made first steps towards bridging the gap between the two approaches, but faced two problems. First, reinforcement learning is typically formulated in a discrete framework, ill-adapted to the description of natural situations. Second, biologically plausible models of reward-modulated spike-timing-dependent plasticity require precise calculation of the reward prediction error, yet it remains to be shown how this can be computed by neurons. Here we propose a solution to these problems by extending the continuous temporal difference (TD) learning of Doya (2000) to the case of spiking neurons in an actor-critic network operating in continuous time, and with continuous state and action representations. In our model, the critic learns to predict expected future rewards in real time. Its activity, together with actual rewards, conditions the delivery of a neuromodulatory TD signal to itself and to the actor, which is responsible for action choice. In simulations, we show that such an architecture can solve a Morris water-maze-like navigation task, in a number of trials consistent with reported animal performance. We also use our model to solve the acrobot and the cartpole problems, two complex motor control tasks. Our model provides a plausible way of computing reward prediction error in the brain. Moreover, the analytically derived learning rule is consistent with experimental evidence for dopamine-modulated spike-timing-dependent plasticity.

Deep Reinforcement Learning of Cell Movement in the Early Stage of C. elegans Embryogenesis.

PubMed

Wang, Zi; Wang, Dali; Li, Chengcheng; Xu, Yichi; Li, Husheng; Bao, Zhirong

2018-04-25

Cell movement in the early phase of C. elegans development is regulated by a highly complex process in which a set of rules and connections are formulated at distinct scales. Previous efforts have demonstrated that agent-based, multi-scale modeling systems can integrate physical and biological rules and provide new avenues to study developmental systems. However, the application of these systems to model cell movement is still challenging and requires a comprehensive understanding of regulatory networks at the right scales. Recent developments in deep learning and reinforcement learning provide an unprecedented opportunity to explore cell movement using 3D time-lapse microscopy images. We present a deep reinforcement learning approach within an agent-based modeling system to characterize cell movement in the embryonic development of C. elegans. Our modeling system captures the complexity of cell movement patterns in the embryo and overcomes the local optimization problem encountered by traditional rule-based, agent-based modeling that uses greedy algorithms. We tested our model with two real developmental processes: the anterior movement of the Cpaaa cell via intercalation and the rearrangement of the superficial left-right asymmetry. In the first case, the model results suggested that Cpaaa's intercalation is an active directional cell movement caused by the continuous effects from a longer distance (farther than the length of two adjacent cells), as opposed to a passive movement caused by neighbor cell movements. In the second case, a leader-follower mechanism well explained the collective cell movement pattern in the asymmetry rearrangement. These results showed that our approach to introduce deep reinforcement learning into agent-based modeling can test regulatory mechanisms by exploring cell migration paths in a reverse engineering perspective. This model opens new doors to explore the large datasets generated by live imaging. Source code is available at https

Electrophysiological correlates of reinforcement learning in young people with Tourette syndrome with and without co-occurring ADHD symptoms.

PubMed

Shephard, Elizabeth; Jackson, Georgina M; Groom, Madeleine J

2016-06-01

Altered reinforcement learning is implicated in the causes of Tourette syndrome (TS) and attention-deficit/hyperactivity disorder (ADHD). TS and ADHD frequently co-occur but how this affects reinforcement learning has not been investigated. We examined the ability of young people with TS (n=18), TS+ADHD (N=17), ADHD (n=13) and typically developing controls (n=20) to learn and reverse stimulus-response (S-R) associations based on positive and negative reinforcement feedback. We used a 2 (TS-yes, TS-no)×2 (ADHD-yes, ADHD-no) factorial design to assess the effects of TS, ADHD, and their interaction on behavioural (accuracy, RT) and event-related potential (stimulus-locked P3, feedback-locked P2, feedback-related negativity, FRN) indices of learning and reversing the S-R associations. TS was associated with intact learning and reversal performance and largely typical ERP amplitudes. ADHD was associated with lower accuracy during S-R learning and impaired reversal learning (significantly reduced accuracy and a trend for smaller P3 amplitude). The results indicate that co-occurring ADHD symptoms impair reversal learning in TS+ADHD. The implications of these findings for behavioural tic therapies are discussed. Copyright © 2016 ISDN. Published by Elsevier Ltd. All rights reserved.

Vicarious reinforcement learning signals when instructing others.

PubMed

Apps, Matthew A J; Lesage, Elise; Ramnani, Narender

2015-02-18

Reinforcement learning (RL) theory posits that learning is driven by discrepancies between the predicted and actual outcomes of actions (prediction errors [PEs]). In social environments, learning is often guided by similar RL mechanisms. For example, teachers monitor the actions of students and provide feedback to them. This feedback evokes PEs in students that guide their learning. We report the first study that investigates the neural mechanisms that underpin RL signals in the brain of a teacher. Neurons in the anterior cingulate cortex (ACC) signal PEs when learning from the outcomes of one's own actions but also signal information when outcomes are received by others. Does a teacher's ACC signal PEs when monitoring a student's learning? Using fMRI, we studied brain activity in human subjects (teachers) as they taught a confederate (student) action-outcome associations by providing positive or negative feedback. We examined activity time-locked to the students' responses, when teachers infer student predictions and know actual outcomes. We fitted a RL-based computational model to the behavior of the student to characterize their learning, and examined whether a teacher's ACC signals when a student's predictions are wrong. In line with our hypothesis, activity in the teacher's ACC covaried with the PE values in the model. Additionally, activity in the teacher's insula and ventromedial prefrontal cortex covaried with the predicted value according to the student. Our findings highlight that the ACC signals PEs vicariously for others' erroneous predictions, when monitoring and instructing their learning. These results suggest that RL mechanisms, processed vicariously, may underpin and facilitate teaching behaviors. Copyright © 2015 Apps et al.

Optimizing microstimulation using a reinforcement learning framework.

PubMed

Brockmeier, Austin J; Choi, John S; Distasio, Marcello M; Francis, Joseph T; Príncipe, José C

2011-01-01

The ability to provide sensory feedback is desired to enhance the functionality of neuroprosthetics. Somatosensory feedback provides closed-loop control to the motor system, which is lacking in feedforward neuroprosthetics. In the case of existing somatosensory function, a template of the natural response can be used as a template of desired response elicited by electrical microstimulation. In the case of no initial training data, microstimulation parameters that produce responses close to the template must be selected in an online manner. We propose using reinforcement learning as a framework to balance the exploration of the parameter space and the continued selection of promising parameters for further stimulation. This approach avoids an explicit model of the neural response from stimulation. We explore a preliminary architecture--treating the task as a k-armed bandit--using offline data recorded for natural touch and thalamic microstimulation, and we examine the methods efficiency in exploring the parameter space while concentrating on promising parameter forms. The best matching stimulation parameters, from k = 68 different forms, are selected by the reinforcement learning algorithm consistently after 334 realizations.

From free energy to expected energy: Improving energy-based value function approximation in reinforcement learning.

PubMed

Elfwing, Stefan; Uchibe, Eiji; Doya, Kenji

2016-12-01

Free-energy based reinforcement learning (FERL) was proposed for learning in high-dimensional state and action spaces. However, the FERL method does only really work well with binary, or close to binary, state input, where the number of active states is fewer than the number of non-active states. In the FERL method, the value function is approximated by the negative free energy of a restricted Boltzmann machine (RBM). In our earlier study, we demonstrated that the performance and the robustness of the FERL method can be improved by scaling the free energy by a constant that is related to the size of network. In this study, we propose that RBM function approximation can be further improved by approximating the value function by the negative expected energy (EERL), instead of the negative free energy, as well as being able to handle continuous state input. We validate our proposed method by demonstrating that EERL: (1) outperforms FERL, as well as standard neural network and linear function approximation, for three versions of a gridworld task with high-dimensional image state input; (2) achieves new state-of-the-art results in stochastic SZ-Tetris in both model-free and model-based learning settings; and (3) significantly outperforms FERL and standard neural network function approximation for a robot navigation task with raw and noisy RGB images as state input and a large number of actions. Copyright © 2016 The Author(s). Published by Elsevier Ltd.. All rights reserved.

Grounding the Meanings in Sensorimotor Behavior using Reinforcement Learning

PubMed Central

Farkaš, Igor; Malík, Tomáš; Rebrová, Kristína

2012-01-01

The recent outburst of interest in cognitive developmental robotics is fueled by the ambition to propose ecologically plausible mechanisms of how, among other things, a learning agent/robot could ground linguistic meanings in its sensorimotor behavior. Along this stream, we propose a model that allows the simulated iCub robot to learn the meanings of actions (point, touch, and push) oriented toward objects in robot’s peripersonal space. In our experiments, the iCub learns to execute motor actions and comment on them. Architecturally, the model is composed of three neural-network-based modules that are trained in different ways. The first module, a two-layer perceptron, is trained by back-propagation to attend to the target position in the visual scene, given the low-level visual information and the feature-based target information. The second module, having the form of an actor-critic architecture, is the most distinguishing part of our model, and is trained by a continuous version of reinforcement learning to execute actions as sequences, based on a linguistic command. The third module, an echo-state network, is trained to provide the linguistic description of the executed actions. The trained model generalizes well in case of novel action-target combinations with randomized initial arm positions. It can also promptly adapt its behavior if the action/target suddenly changes during motor execution. PMID:22393319

Working memory contributions to reinforcement learning impairments in schizophrenia.

PubMed

Collins, Anne G E; Brown, Jaime K; Gold, James M; Waltz, James A; Frank, Michael J

2014-10-08

Previous research has shown that patients with schizophrenia are impaired in reinforcement learning tasks. However, behavioral learning curves in such tasks originate from the interaction of multiple neural processes, including the basal ganglia- and dopamine-dependent reinforcement learning (RL) system, but also prefrontal cortex-dependent cognitive strategies involving working memory (WM). Thus, it is unclear which specific system induces impairments in schizophrenia. We recently developed a task and computational model allowing us to separately assess the roles of RL (slow, cumulative learning) mechanisms versus WM (fast but capacity-limited) mechanisms in healthy adult human subjects. Here, we used this task to assess patients' specific sources of impairments in learning. In 15 separate blocks, subjects learned to pick one of three actions for stimuli. The number of stimuli to learn in each block varied from two to six, allowing us to separate influences of capacity-limited WM from the incremental RL system. As expected, both patients (n = 49) and healthy controls (n = 36) showed effects of set size and delay between stimulus repetitions, confirming the presence of working memory effects. Patients performed significantly worse than controls overall, but computational model fits and behavioral analyses indicate that these deficits could be entirely accounted for by changes in WM parameters (capacity and reliability), whereas RL processes were spared. These results suggest that the working memory system contributes strongly to learning impairments in schizophrenia. Copyright © 2014 the authors 0270-6474/14/3413747-10$15.00/0.

Working Memory Contributions to Reinforcement Learning Impairments in Schizophrenia

PubMed Central

Brown, Jaime K.; Gold, James M.; Waltz, James A.; Frank, Michael J.

2014-01-01

Previous research has shown that patients with schizophrenia are impaired in reinforcement learning tasks. However, behavioral learning curves in such tasks originate from the interaction of multiple neural processes, including the basal ganglia- and dopamine-dependent reinforcement learning (RL) system, but also prefrontal cortex-dependent cognitive strategies involving working memory (WM). Thus, it is unclear which specific system induces impairments in schizophrenia. We recently developed a task and computational model allowing us to separately assess the roles of RL (slow, cumulative learning) mechanisms versus WM (fast but capacity-limited) mechanisms in healthy adult human subjects. Here, we used this task to assess patients' specific sources of impairments in learning. In 15 separate blocks, subjects learned to pick one of three actions for stimuli. The number of stimuli to learn in each block varied from two to six, allowing us to separate influences of capacity-limited WM from the incremental RL system. As expected, both patients (n = 49) and healthy controls (n = 36) showed effects of set size and delay between stimulus repetitions, confirming the presence of working memory effects. Patients performed significantly worse than controls overall, but computational model fits and behavioral analyses indicate that these deficits could be entirely accounted for by changes in WM parameters (capacity and reliability), whereas RL processes were spared. These results suggest that the working memory system contributes strongly to learning impairments in schizophrenia. PMID:25297101

Credit assignment in movement-dependent reinforcement learning

PubMed Central

Boggess, Matthew J.; Crossley, Matthew J.; Parvin, Darius; Ivry, Richard B.; Taylor, Jordan A.

2016-01-01

When a person fails to obtain an expected reward from an object in the environment, they face a credit assignment problem: Did the absence of reward reflect an extrinsic property of the environment or an intrinsic error in motor execution? To explore this problem, we modified a popular decision-making task used in studies of reinforcement learning, the two-armed bandit task. We compared a version in which choices were indicated by key presses, the standard response in such tasks, to a version in which the choices were indicated by reaching movements, which affords execution failures. In the key press condition, participants exhibited a strong risk aversion bias; strikingly, this bias reversed in the reaching condition. This result can be explained by a reinforcement model wherein movement errors influence decision-making, either by gating reward prediction errors or by modifying an implicit representation of motor competence. Two further experiments support the gating hypothesis. First, we used a condition in which we provided visual cues indicative of movement errors but informed the participants that trial outcomes were independent of their actual movements. The main result was replicated, indicating that the gating process is independent of participants’ explicit sense of control. Second, individuals with cerebellar degeneration failed to modulate their behavior between the key press and reach conditions, providing converging evidence of an implicit influence of movement error signals on reinforcement learning. These results provide a mechanistically tractable solution to the credit assignment problem. PMID:27247404

Credit assignment in movement-dependent reinforcement learning.

PubMed

McDougle, Samuel D; Boggess, Matthew J; Crossley, Matthew J; Parvin, Darius; Ivry, Richard B; Taylor, Jordan A

2016-06-14

When a person fails to obtain an expected reward from an object in the environment, they face a credit assignment problem: Did the absence of reward reflect an extrinsic property of the environment or an intrinsic error in motor execution? To explore this problem, we modified a popular decision-making task used in studies of reinforcement learning, the two-armed bandit task. We compared a version in which choices were indicated by key presses, the standard response in such tasks, to a version in which the choices were indicated by reaching movements, which affords execution failures. In the key press condition, participants exhibited a strong risk aversion bias; strikingly, this bias reversed in the reaching condition. This result can be explained by a reinforcement model wherein movement errors influence decision-making, either by gating reward prediction errors or by modifying an implicit representation of motor competence. Two further experiments support the gating hypothesis. First, we used a condition in which we provided visual cues indicative of movement errors but informed the participants that trial outcomes were independent of their actual movements. The main result was replicated, indicating that the gating process is independent of participants' explicit sense of control. Second, individuals with cerebellar degeneration failed to modulate their behavior between the key press and reach conditions, providing converging evidence of an implicit influence of movement error signals on reinforcement learning. These results provide a mechanistically tractable solution to the credit assignment problem.

Apprenticeship Learning: Learning to Schedule from Human Experts

DTIC Science & Technology

2016-06-09

approaches to learning such models are based on Markov models, such as reinforcement learning or inverse reinforcement learning (Busoniu, Babuska, and De...via inverse reinforcement learning. In ICML. Barto, A. G., and Mahadevan, S. 2003. Recent advances in hierarchical reinforcement learning. Discrete...of tasks with temporal constraints. In Proc. AAAI, 2110–2116. Odom, P., and Natarajan, S. 2015. Active advice seeking for inverse reinforcement

Deep reinforcement learning for automated radiation adaptation in lung cancer.

PubMed

Tseng, Huan-Hsin; Luo, Yi; Cui, Sunan; Chien, Jen-Tzung; Ten Haken, Randall K; Naqa, Issam El

2017-12-01

To investigate deep reinforcement learning (DRL) based on historical treatment plans for developing automated radiation adaptation protocols for nonsmall cell lung cancer (NSCLC) patients that aim to maximize tumor local control at reduced rates of radiation pneumonitis grade 2 (RP2). In a retrospective population of 114 NSCLC patients who received radiotherapy, a three-component neural networks framework was developed for deep reinforcement learning (DRL) of dose fractionation adaptation. Large-scale patient characteristics included clinical, genetic, and imaging radiomics features in addition to tumor and lung dosimetric variables. First, a generative adversarial network (GAN) was employed to learn patient population characteristics necessary for DRL training from a relatively limited sample size. Second, a radiotherapy artificial environment (RAE) was reconstructed by a deep neural network (DNN) utilizing both original and synthetic data (by GAN) to estimate the transition probabilities for adaptation of personalized radiotherapy patients' treatment courses. Third, a deep Q-network (DQN) was applied to the RAE for choosing the optimal dose in a response-adapted treatment setting. This multicomponent reinforcement learning approach was benchmarked against real clinical decisions that were applied in an adaptive dose escalation clinical protocol. In which, 34 patients were treated based on avid PET signal in the tumor and constrained by a 17.2% normal tissue complication probability (NTCP) limit for RP2. The uncomplicated cure probability (P+) was used as a baseline reward function in the DRL. Taking our adaptive dose escalation protocol as a blueprint for the proposed DRL (GAN + RAE + DQN) architecture, we obtained an automated dose adaptation estimate for use at ∼2/3 of the way into the radiotherapy treatment course. By letting the DQN component freely control the estimated adaptive dose per fraction (ranging from 1-5 Gy), the DRL automatically favored dose

Reconciling Reinforcement Learning Models with Behavioral Extinction and Renewal: Implications for Addiction, Relapse, and Problem Gambling

ERIC Educational Resources Information Center

Redish, A. David; Jensen, Steve; Johnson, Adam; Kurth-Nelson, Zeb

2007-01-01

Because learned associations are quickly renewed following extinction, the extinction process must include processes other than unlearning. However, reinforcement learning models, such as the temporal difference reinforcement learning (TDRL) model, treat extinction as an unlearning of associated value and are thus unable to capture renewal. TDRL…

Functional Contour-following via Haptic Perception and Reinforcement Learning.

PubMed

Hellman, Randall B; Tekin, Cem; van der Schaar, Mihaela; Santos, Veronica J

2018-01-01

Many tasks involve the fine manipulation of objects despite limited visual feedback. In such scenarios, tactile and proprioceptive feedback can be leveraged for task completion. We present an approach for real-time haptic perception and decision-making for a haptics-driven, functional contour-following task: the closure of a ziplock bag. This task is challenging for robots because the bag is deformable, transparent, and visually occluded by artificial fingertip sensors that are also compliant. A deep neural net classifier was trained to estimate the state of a zipper within a robot's pinch grasp. A Contextual Multi-Armed Bandit (C-MAB) reinforcement learning algorithm was implemented to maximize cumulative rewards by balancing exploration versus exploitation of the state-action space. The C-MAB learner outperformed a benchmark Q-learner by more efficiently exploring the state-action space while learning a hard-to-code task. The learned C-MAB policy was tested with novel ziplock bag scenarios and contours (wire, rope). Importantly, this work contributes to the development of reinforcement learning approaches that account for limited resources such as hardware life and researcher time. As robots are used to perform complex, physically interactive tasks in unstructured or unmodeled environments, it becomes important to develop methods that enable efficient and effective learning with physical testbeds.

Identifying Cognitive Remediation Change Through Computational Modelling—Effects on Reinforcement Learning in Schizophrenia

PubMed Central

Cella, Matteo; Bishara, Anthony J.; Medin, Evelina; Swan, Sarah; Reeder, Clare; Wykes, Til

2014-01-01

Objective: Converging research suggests that individuals with schizophrenia show a marked impairment in reinforcement learning, particularly in tasks requiring flexibility and adaptation. The problem has been associated with dopamine reward systems. This study explores, for the first time, the characteristics of this impairment and how it is affected by a behavioral intervention—cognitive remediation. Method: Using computational modelling, 3 reinforcement learning parameters based on the Wisconsin Card Sorting Test (WCST) trial-by-trial performance were estimated: R (reward sensitivity), P (punishment sensitivity), and D (choice consistency). In Study 1 the parameters were compared between a group of individuals with schizophrenia (n = 100) and a healthy control group (n = 50). In Study 2 the effect of cognitive remediation therapy (CRT) on these parameters was assessed in 2 groups of individuals with schizophrenia, one receiving CRT (n = 37) and the other receiving treatment as usual (TAU, n = 34). Results: In Study 1 individuals with schizophrenia showed impairment in the R and P parameters compared with healthy controls. Study 2 demonstrated that sensitivity to negative feedback (P) and reward (R) improved in the CRT group after therapy compared with the TAU group. R and P parameter change correlated with WCST outputs. Improvements in R and P after CRT were associated with working memory gains and reduction of negative symptoms, respectively. Conclusion: Schizophrenia reinforcement learning difficulties negatively influence performance in shift learning tasks. CRT can improve sensitivity to reward and punishment. Identifying parameters that show change may be useful in experimental medicine studies to identify cognitive domains susceptible to improvement. PMID:24214932

Learning to use working memory: a reinforcement learning gating model of rule acquisition in rats

PubMed Central

Lloyd, Kevin; Becker, Nadine; Jones, Matthew W.; Bogacz, Rafal

2012-01-01

Learning to form appropriate, task-relevant working memory representations is a complex process central to cognition. Gating models frame working memory as a collection of past observations and use reinforcement learning (RL) to solve the problem of when to update these observations. Investigation of how gating models relate to brain and behavior remains, however, at an early stage. The current study sought to explore the ability of simple RL gating models to replicate rule learning behavior in rats. Rats were trained in a maze-based spatial learning task that required animals to make trial-by-trial choices contingent upon their previous experience. Using an abstract version of this task, we tested the ability of two gating algorithms, one based on the Actor-Critic and the other on the State-Action-Reward-State-Action (SARSA) algorithm, to generate behavior consistent with the rats'. Both models produced rule-acquisition behavior consistent with the experimental data, though only the SARSA gating model mirrored faster learning following rule reversal. We also found that both gating models learned multiple strategies in solving the initial task, a property which highlights the multi-agent nature of such models and which is of importance in considering the neural basis of individual differences in behavior. PMID:23115551

A common neural circuit mechanism for internally guided and externally reinforced forms of motor learning.

PubMed

Hisey, Erin; Kearney, Matthew Gene; Mooney, Richard

2018-04-01

The complex skills underlying verbal and musical expression can be learned without external punishment or reward, indicating their learning is internally guided. The neural mechanisms that mediate internally guided learning are poorly understood, but a circuit comprising dopamine-releasing neurons in the midbrain ventral tegmental area (VTA) and their targets in the basal ganglia are important to externally reinforced learning. Juvenile zebra finches copy a tutor song in a process that is internally guided and, in adulthood, can learn to modify the fundamental frequency (pitch) of a target syllable in response to external reinforcement with white noise. Here we combined intersectional genetic ablation of VTA neurons, reversible blockade of dopamine receptors in the basal ganglia, and singing-triggered optogenetic stimulation of VTA terminals to establish that a common VTA-basal ganglia circuit enables internally guided song copying and externally reinforced syllable pitch learning.

Challenges in the Verification of Reinforcement Learning Algorithms

NASA Technical Reports Server (NTRS)

Van Wesel, Perry; Goodloe, Alwyn E.

2017-01-01

Machine learning (ML) is increasingly being applied to a wide array of domains from search engines to autonomous vehicles. These algorithms, however, are notoriously complex and hard to verify. This work looks at the assumptions underlying machine learning algorithms as well as some of the challenges in trying to verify ML algorithms. Furthermore, we focus on the specific challenges of verifying reinforcement learning algorithms. These are highlighted using a specific example. Ultimately, we do not offer a solution to the complex problem of ML verification, but point out possible approaches for verification and interesting research opportunities.

Reinforcement learning of targeted movement in a spiking neuronal model of motor cortex.

PubMed

Chadderdon, George L; Neymotin, Samuel A; Kerr, Cliff C; Lytton, William W

2012-01-01

Sensorimotor control has traditionally been considered from a control theory perspective, without relation to neurobiology. In contrast, here we utilized a spiking-neuron model of motor cortex and trained it to perform a simple movement task, which consisted of rotating a single-joint "forearm" to a target. Learning was based on a reinforcement mechanism analogous to that of the dopamine system. This provided a global reward or punishment signal in response to decreasing or increasing distance from hand to target, respectively. Output was partially driven by Poisson motor babbling, creating stochastic movements that could then be shaped by learning. The virtual forearm consisted of a single segment rotated around an elbow joint, controlled by flexor and extensor muscles. The model consisted of 144 excitatory and 64 inhibitory event-based neurons, each with AMPA, NMDA, and GABA synapses. Proprioceptive cell input to this model encoded the 2 muscle lengths. Plasticity was only enabled in feedforward connections between input and output excitatory units, using spike-timing-dependent eligibility traces for synaptic credit or blame assignment. Learning resulted from a global 3-valued signal: reward (+1), no learning (0), or punishment (-1), corresponding to phasic increases, lack of change, or phasic decreases of dopaminergic cell firing, respectively. Successful learning only occurred when both reward and punishment were enabled. In this case, 5 target angles were learned successfully within 180 s of simulation time, with a median error of 8 degrees. Motor babbling allowed exploratory learning, but decreased the stability of the learned behavior, since the hand continued moving after reaching the target. Our model demonstrated that a global reinforcement signal, coupled with eligibility traces for synaptic plasticity, can train a spiking sensorimotor network to perform goal-directed motor behavior.

Social Learning, Reinforcement and Crime: Evidence from Three European Cities

ERIC Educational Resources Information Center

Tittle, Charles R.; Antonaccio, Olena; Botchkovar, Ekaterina

2012-01-01

This study reports a cross-cultural test of Social Learning Theory using direct measures of social learning constructs and focusing on the causal structure implied by the theory. Overall, the results strongly confirm the main thrust of the theory. Prior criminal reinforcement and current crime-favorable definitions are highly related in all three…

Modeling Avoidance in Mood and Anxiety Disorders Using Reinforcement Learning.

PubMed

Mkrtchian, Anahit; Aylward, Jessica; Dayan, Peter; Roiser, Jonathan P; Robinson, Oliver J

2017-10-01

Serious and debilitating symptoms of anxiety are the most common mental health problem worldwide, accounting for around 5% of all adult years lived with disability in the developed world. Avoidance behavior-avoiding social situations for fear of embarrassment, for instance-is a core feature of such anxiety. However, as for many other psychiatric symptoms the biological mechanisms underlying avoidance remain unclear. Reinforcement learning models provide formal and testable characterizations of the mechanisms of decision making; here, we examine avoidance in these terms. A total of 101 healthy participants and individuals with mood and anxiety disorders completed an approach-avoidance go/no-go task under stress induced by threat of unpredictable shock. We show an increased reliance in the mood and anxiety group on a parameter of our reinforcement learning model that characterizes a prepotent (pavlovian) bias to withhold responding in the face of negative outcomes. This was particularly the case when the mood and anxiety group was under stress. This formal description of avoidance within the reinforcement learning framework provides a new means of linking clinical symptoms with biophysically plausible models of neural circuitry and, as such, takes us closer to a mechanistic understanding of mood and anxiety disorders. Copyright © 2017 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved.

Kinesthetic Reinforcement-Is It a Boon to Learning?

ERIC Educational Resources Information Center

Bohrer, Roxilu K.

1970-01-01

Language instruction, particularly in the elementary school, should be reinforced through the use of visual aids and through associated physical activity. Kinesthetic experiences provide an opportunity to make use of non-verbal cues to meaning, enliven classroom activities, and maximize learning for pupils. The author discusses the educational…

Reinforcement Learning in a Nonstationary Environment: The El Farol Problem

NASA Technical Reports Server (NTRS)

Bell, Ann Maria

1999-01-01

This paper examines the performance of simple learning rules in a complex adaptive system based on a coordination problem modeled on the El Farol problem. The key features of the El Farol problem are that it typically involves a medium number of agents and that agents' pay-off functions have a discontinuous response to increased congestion. First we consider a single adaptive agent facing a stationary environment. We demonstrate that the simple learning rules proposed by Roth and Er'ev can be extremely sensitive to small changes in the initial conditions and that events early in a simulation can affect the performance of the rule over a relatively long time horizon. In contrast, a reinforcement learning rule based on standard practice in the computer science literature converges rapidly and robustly. The situation is reversed when multiple adaptive agents interact: the RE algorithms often converge rapidly to a stable average aggregate attendance despite the slow and erratic behavior of individual learners, while the CS based learners frequently over-attend in the early and intermediate terms. The symmetric mixed strategy equilibria is unstable: all three learning rules ultimately tend towards pure strategies or stabilize in the medium term at non-equilibrium probabilities of attendance. The brittleness of the algorithms in different contexts emphasize the importance of thorough and thoughtful examination of simulation-based results.

Applying reinforcement learning techniques to detect hepatocellular carcinoma under limited screening capacity.

PubMed

Lee, Elliot; Lavieri, Mariel S; Volk, Michael L; Xu, Yongcai

2015-09-01

We investigate the problem faced by a healthcare system wishing to allocate its constrained screening resources across a population at risk for developing a disease. A patient's risk of developing the disease depends on his/her biomedical dynamics. However, knowledge of these dynamics must be learned by the system over time. Three classes of reinforcement learning policies are designed to address this problem of simultaneously gathering and utilizing information across multiple patients. We investigate a case study based upon the screening for Hepatocellular Carcinoma (HCC), and optimize each of the three classes of policies using the indifference zone method. A simulation is built to gauge the performance of these policies, and their performance is compared to current practice. We then demonstrate how the benefits of learning-based screening policies differ across various levels of resource scarcity and provide metrics of policy performance.

Intrinsically motivated reinforcement learning for human-robot interaction in the real-world.

PubMed

Qureshi, Ahmed Hussain; Nakamura, Yutaka; Yoshikawa, Yuichiro; Ishiguro, Hiroshi

2018-03-26

For a natural social human-robot interaction, it is essential for a robot to learn the human-like social skills. However, learning such skills is notoriously hard due to the limited availability of direct instructions from people to teach a robot. In this paper, we propose an intrinsically motivated reinforcement learning framework in which an agent gets the intrinsic motivation-based rewards through the action-conditional predictive model. By using the proposed method, the robot learned the social skills from the human-robot interaction experiences gathered in the real uncontrolled environments. The results indicate that the robot not only acquired human-like social skills but also took more human-like decisions, on a test dataset, than a robot which received direct rewards for the task achievement. Copyright © 2018 Elsevier Ltd. All rights reserved.

Design issues of a reinforcement-based self-learning fuzzy controller for petrochemical process control

NASA Technical Reports Server (NTRS)

Yen, John; Wang, Haojin; Daugherity, Walter C.

1992-01-01

Fuzzy logic controllers have some often-cited advantages over conventional techniques such as PID control, including easier implementation, accommodation to natural language, and the ability to cover a wider range of operating conditions. One major obstacle that hinders the broader application of fuzzy logic controllers is the lack of a systematic way to develop and modify their rules; as a result the creation and modification of fuzzy rules often depends on trial and error or pure experimentation. One of the proposed approaches to address this issue is a self-learning fuzzy logic controller (SFLC) that uses reinforcement learning techniques to learn the desirability of states and to adjust the consequent part of its fuzzy control rules accordingly. Due to the different dynamics of the controlled processes, the performance of a self-learning fuzzy controller is highly contingent on its design. The design issue has not received sufficient attention. The issues related to the design of a SFLC for application to a petrochemical process are discussed, and its performance is compared with that of a PID and a self-tuning fuzzy logic controller.

Neural Control of a Tracking Task via Attention-Gated Reinforcement Learning for Brain-Machine Interfaces.

PubMed

Wang, Yiwen; Wang, Fang; Xu, Kai; Zhang, Qiaosheng; Zhang, Shaomin; Zheng, Xiaoxiang

2015-05-01

Reinforcement learning (RL)-based brain machine interfaces (BMIs) enable the user to learn from the environment through interactions to complete the task without desired signals, which is promising for clinical applications. Previous studies exploited Q-learning techniques to discriminate neural states into simple directional actions providing the trial initial timing. However, the movements in BMI applications can be quite complicated, and the action timing explicitly shows the intention when to move. The rich actions and the corresponding neural states form a large state-action space, imposing generalization difficulty on Q-learning. In this paper, we propose to adopt attention-gated reinforcement learning (AGREL) as a new learning scheme for BMIs to adaptively decode high-dimensional neural activities into seven distinct movements (directional moves, holdings and resting) due to the efficient weight-updating. We apply AGREL on neural data recorded from M1 of a monkey to directly predict a seven-action set in a time sequence to reconstruct the trajectory of a center-out task. Compared to Q-learning techniques, AGREL could improve the target acquisition rate to 90.16% in average with faster convergence and more stability to follow neural activity over multiple days, indicating the potential to achieve better online decoding performance for more complicated BMI tasks.

Learning the specific quality of taste reinforcement in larval Drosophila.

PubMed

Schleyer, Michael; Miura, Daisuke; Tanimura, Teiichi; Gerber, Bertram

2015-01-27

The only property of reinforcement insects are commonly thought to learn about is its value. We show that larval Drosophila not only remember the value of reinforcement (How much?), but also its quality (What?). This is demonstrated both within the appetitive domain by using sugar vs amino acid as different reward qualities, and within the aversive domain by using bitter vs high-concentration salt as different qualities of punishment. From the available literature, such nuanced memories for the quality of reinforcement are unexpected and pose a challenge to present models of how insect memory is organized. Given that animals as simple as larval Drosophila, endowed with but 10,000 neurons, operate with both reinforcement value and quality, we suggest that both are fundamental aspects of mnemonic processing-in any brain.

Cost-Benefit Arbitration Between Multiple Reinforcement-Learning Systems.

PubMed

Kool, Wouter; Gershman, Samuel J; Cushman, Fiery A

2017-09-01

Human behavior is sometimes determined by habit and other times by goal-directed planning. Modern reinforcement-learning theories formalize this distinction as a competition between a computationally cheap but inaccurate model-free system that gives rise to habits and a computationally expensive but accurate model-based system that implements planning. It is unclear, however, how people choose to allocate control between these systems. Here, we propose that arbitration occurs by comparing each system's task-specific costs and benefits. To investigate this proposal, we conducted two experiments showing that people increase model-based control when it achieves greater accuracy than model-free control, and especially when the rewards of accurate performance are amplified. In contrast, they are insensitive to reward amplification when model-based and model-free control yield equivalent accuracy. This suggests that humans adaptively balance habitual and planned action through on-line cost-benefit analysis.

Classification of amyotrophic lateral sclerosis disease based on convolutional neural network and reinforcement sample learning algorithm.

PubMed

Sengur, Abdulkadir; Akbulut, Yaman; Guo, Yanhui; Bajaj, Varun

2017-12-01

Electromyogram (EMG) signals contain useful information of the neuromuscular diseases like amyotrophic lateral sclerosis (ALS). ALS is a well-known brain disease, which can progressively degenerate the motor neurons. In this paper, we propose a deep learning based method for efficient classification of ALS and normal EMG signals. Spectrogram, continuous wavelet transform (CWT), and smoothed pseudo Wigner-Ville distribution (SPWVD) have been employed for time-frequency (T-F) representation of EMG signals. A convolutional neural network is employed to classify these features. In it, Two convolution layers, two pooling layer, a fully connected layer and a lost function layer is considered in CNN architecture. The CNN architecture is trained with the reinforcement sample learning strategy. The efficiency of the proposed implementation is tested on publicly available EMG dataset. The dataset contains 89 ALS and 133 normal EMG signals with 24 kHz sampling frequency. Experimental results show 96.80% accuracy. The obtained results are also compared with other methods, which show the superiority of the proposed method.

The effects of aging on the interaction between reinforcement learning and attention.

PubMed

Radulescu, Angela; Daniel, Reka; Niv, Yael

2016-11-01

Reinforcement learning (RL) in complex environments relies on selective attention to uncover those aspects of the environment that are most predictive of reward. Whereas previous work has focused on age-related changes in RL, it is not known whether older adults learn differently from younger adults when selective attention is required. In 2 experiments, we examined how aging affects the interaction between RL and selective attention. Younger and older adults performed a learning task in which only 1 stimulus dimension was relevant to predicting reward, and within it, 1 "target" feature was the most rewarding. Participants had to discover this target feature through trial and error. In Experiment 1, stimuli varied on 1 or 3 dimensions and participants received hints that revealed the target feature, the relevant dimension, or gave no information. Group-related differences in accuracy and RTs differed systematically as a function of the number of dimensions and the type of hint available. In Experiment 2 we used trial-by-trial computational modeling of the learning process to test for age-related differences in learning strategies. Behavior of both young and older adults was explained well by a reinforcement-learning model that uses selective attention to constrain learning. However, the model suggested that older adults restricted their learning to fewer features, employing more focused attention than younger adults. Furthermore, this difference in strategy predicted age-related deficits in accuracy. We discuss these results suggesting that a narrower filter of attention may reflect an adaptation to the reduced capabilities of the reinforcement learning system. (PsycINFO Database Record (c) 2016 APA, all rights reserved).

Intelligent multiagent coordination based on reinforcement hierarchical neuro-fuzzy models.

PubMed

Mendoza, Leonardo Forero; Vellasco, Marley; Figueiredo, Karla

2014-12-01

This paper presents the research and development of two hybrid neuro-fuzzy models for the hierarchical coordination of multiple intelligent agents. The main objective of the models is to have multiple agents interact intelligently with each other in complex systems. We developed two new models of coordination for intelligent multiagent systems, which integrates the Reinforcement Learning Hierarchical Neuro-Fuzzy model with two proposed coordination mechanisms: the MultiAgent Reinforcement Learning Hierarchical Neuro-Fuzzy with a market-driven coordination mechanism (MA-RL-HNFP-MD) and the MultiAgent Reinforcement Learning Hierarchical Neuro-Fuzzy with graph coordination (MA-RL-HNFP-CG). In order to evaluate the proposed models and verify the contribution of the proposed coordination mechanisms, two multiagent benchmark applications were developed: the pursuit game and the robot soccer simulation. The results obtained demonstrated that the proposed coordination mechanisms greatly improve the performance of the multiagent system when compared with other strategies.

Learning the specific quality of taste reinforcement in larval Drosophila

PubMed Central

Schleyer, Michael; Miura, Daisuke; Tanimura, Teiichi; Gerber, Bertram

2015-01-01

The only property of reinforcement insects are commonly thought to learn about is its value. We show that larval Drosophila not only remember the value of reinforcement (How much?), but also its quality (What?). This is demonstrated both within the appetitive domain by using sugar vs amino acid as different reward qualities, and within the aversive domain by using bitter vs high-concentration salt as different qualities of punishment. From the available literature, such nuanced memories for the quality of reinforcement are unexpected and pose a challenge to present models of how insect memory is organized. Given that animals as simple as larval Drosophila, endowed with but 10,000 neurons, operate with both reinforcement value and quality, we suggest that both are fundamental aspects of mnemonic processing—in any brain. DOI: http://dx.doi.org/10.7554/eLife.04711.001 PMID:25622533

Reinforcement learning design-based adaptive tracking control with less learning parameters for nonlinear discrete-time MIMO systems.

PubMed

Liu, Yan-Jun; Tang, Li; Tong, Shaocheng; Chen, C L Philip; Li, Dong-Juan

2015-01-01

Based on the neural network (NN) approximator, an online reinforcement learning algorithm is proposed for a class of affine multiple input and multiple output (MIMO) nonlinear discrete-time systems with unknown functions and disturbances. In the design procedure, two networks are provided where one is an action network to generate an optimal control signal and the other is a critic network to approximate the cost function. An optimal control signal and adaptation laws can be generated based on two NNs. In the previous approaches, the weights of critic and action networks are updated based on the gradient descent rule and the estimations of optimal weight vectors are directly adjusted in the design. Consequently, compared with the existing results, the main contributions of this paper are: 1) only two parameters are needed to be adjusted, and thus the number of the adaptation laws is smaller than the previous results and 2) the updating parameters do not depend on the number of the subsystems for MIMO systems and the tuning rules are replaced by adjusting the norms on optimal weight vectors in both action and critic networks. It is proven that the tracking errors, the adaptation laws, and the control inputs are uniformly bounded using Lyapunov analysis method. The simulation examples are employed to illustrate the effectiveness of the proposed algorithm.

A new computational account of cognitive control over reinforcement-based decision-making: Modeling of a probabilistic learning task.

PubMed

Zendehrouh, Sareh

2015-11-01

Recent work on decision-making field offers an account of dual-system theory for decision-making process. This theory holds that this process is conducted by two main controllers: a goal-directed system and a habitual system. In the reinforcement learning (RL) domain, the habitual behaviors are connected with model-free methods, in which appropriate actions are learned through trial-and-error experiences. However, goal-directed behaviors are associated with model-based methods of RL, in which actions are selected using a model of the environment. Studies on cognitive control also suggest that during processes like decision-making, some cortical and subcortical structures work in concert to monitor the consequences of decisions and to adjust control according to current task demands. Here a computational model is presented based on dual system theory and cognitive control perspective of decision-making. The proposed model is used to simulate human performance on a variant of probabilistic learning task. The basic proposal is that the brain implements a dual controller, while an accompanying monitoring system detects some kinds of conflict including a hypothetical cost-conflict one. The simulation results address existing theories about two event-related potentials, namely error related negativity (ERN) and feedback related negativity (FRN), and explore the best account of them. Based on the results, some testable predictions are also presented. Copyright © 2015 Elsevier Ltd. All rights reserved.

Generalization of value in reinforcement learning by humans

PubMed Central

Wimmer, G. Elliott; Daw, Nathaniel D.; Shohamy, Daphna

2012-01-01

Research in decision making has focused on the role of dopamine and its striatal targets in guiding choices via learned stimulus-reward or stimulus-response associations, behavior that is well-described by reinforcement learning (RL) theories. However, basic RL is relatively limited in scope and does not explain how learning about stimulus regularities or relations may guide decision making. A candidate mechanism for this type of learning comes from the domain of memory, which has highlighted a role for the hippocampus in learning of stimulus-stimulus relations, typically dissociated from the role of the striatum in stimulus-response learning. Here, we used fMRI and computational model-based analyses to examine the joint contributions of these mechanisms to RL. Humans performed an RL task with added relational structure, modeled after tasks used to isolate hippocampal contributions to memory. On each trial participants chose one of four options, but the reward probabilities for pairs of options were correlated across trials. This (uninstructed) relationship between pairs of options potentially enabled an observer to learn about options’ values based on experience with the other options and to generalize across them. We observed BOLD activity related to learning in the striatum and also in the hippocampus. By comparing a basic RL model to one augmented to allow feedback to generalize between correlated options, we tested whether choice behavior and BOLD activity were influenced by the opportunity to generalize across correlated options. Although such generalization goes beyond standard computational accounts of RL and striatal BOLD, both choices and striatal BOLD were better explained by the augmented model. Consistent with the hypothesized role for the hippocampus in this generalization, functional connectivity between the ventral striatum and hippocampus was modulated, across participants, by the ability of the augmented model to capture participants’ choice

Measuring reinforcement learning and motivation constructs in experimental animals: relevance to the negative symptoms of schizophrenia.

PubMed

Markou, Athina; Salamone, John D; Bussey, Timothy J; Mar, Adam C; Brunner, Daniela; Gilmour, Gary; Balsam, Peter

2013-11-01

The present review article summarizes and expands upon the discussions that were initiated during a meeting of the Cognitive Neuroscience Treatment Research to Improve Cognition in Schizophrenia (CNTRICS; http://cntrics.ucdavis.edu) meeting. A major goal of the CNTRICS meeting was to identify experimental procedures and measures that can be used in laboratory animals to assess psychological constructs that are related to the psychopathology of schizophrenia. The issues discussed in this review reflect the deliberations of the Motivation Working Group of the CNTRICS meeting, which included most of the authors of this article as well as additional participants. After receiving task nominations from the general research community, this working group was asked to identify experimental procedures in laboratory animals that can assess aspects of reinforcement learning and motivation that may be relevant for research on the negative symptoms of schizophrenia, as well as other disorders characterized by deficits in reinforcement learning and motivation. The tasks described here that assess reinforcement learning are the Autoshaping Task, Probabilistic Reward Learning Tasks, and the Response Bias Probabilistic Reward Task. The tasks described here that assess motivation are Outcome Devaluation and Contingency Degradation Tasks and Effort-Based Tasks. In addition to describing such methods and procedures, the present article provides a working vocabulary for research and theory in this field, as well as an industry perspective about how such tasks may be used in drug discovery. It is hoped that this review can aid investigators who are conducting research in this complex area, promote translational studies by highlighting shared research goals and fostering a common vocabulary across basic and clinical fields, and facilitate the development of medications for the treatment of symptoms mediated by reinforcement learning and motivational deficits. Copyright © 2013 Elsevier

Measuring reinforcement learning and motivation constructs in experimental animals: relevance to the negative symptoms of schizophrenia

PubMed Central

Markou, Athina; Salamone, John D.; Bussey, Timothy; Mar, Adam; Brunner, Daniela; Gilmour, Gary; Balsam, Peter

2013-01-01

The present review article summarizes and expands upon the discussions that were initiated during a meeting of the Cognitive Neuroscience Treatment Research to Improve Cognition in Schizophrenia (CNTRICS; http://cntrics.ucdavis.edu). A major goal of the CNTRICS meeting was to identify experimental procedures and measures that can be used in laboratory animals to assess psychological constructs that are related to the psychopathology of schizophrenia. The issues discussed in this review reflect the deliberations of the Motivation Working Group of the CNTRICS meeting, which included most of the authors of this article as well as additional participants. After receiving task nominations from the general research community, this working group was asked to identify experimental procedures in laboratory animals that can assess aspects of reinforcement learning and motivation that may be relevant for research on the negative symptoms of schizophrenia, as well as other disorders characterized by deficits in reinforcement learning and motivation. The tasks described here that assess reinforcement learning are the Autoshaping Task, Probabilistic Reward Learning Tasks, and the Response Bias Probabilistic Reward Task. The tasks described here that assess motivation are Outcome Devaluation and Contingency Degradation Tasks and Effort-Based Tasks. In addition to describing such methods and procedures, the present article provides a working vocabulary for research and theory in this field, as well as an industry perspective about how such tasks may be used in drug discovery. It is hoped that this review can aid investigators who are conducting research in this complex area, promote translational studies by highlighting shared research goals and fostering a common vocabulary across basic and clinical fields, and facilitate the development of medications for the treatment of symptoms mediated by reinforcement learning and motivational deficits. PMID:23994273

A reinforcement learning model of joy, distress, hope and fear

NASA Astrophysics Data System (ADS)

Broekens, Joost; Jacobs, Elmer; Jonker, Catholijn M.

2015-07-01

In this paper we computationally study the relation between adaptive behaviour and emotion. Using the reinforcement learning framework, we propose that learned state utility, ?, models fear (negative) and hope (positive) based on the fact that both signals are about anticipation of loss or gain. Further, we propose that joy/distress is a signal similar to the error signal. We present agent-based simulation experiments that show that this model replicates psychological and behavioural dynamics of emotion. This work distinguishes itself by assessing the dynamics of emotion in an adaptive agent framework - coupling it to the literature on habituation, development, extinction and hope theory. Our results support the idea that the function of emotion is to provide a complex feedback signal for an organism to adapt its behaviour. Our work is relevant for understanding the relation between emotion and adaptation in animals, as well as for human-robot interaction, in particular how emotional signals can be used to communicate between adaptive agents and humans.

Goal-Directed and Habit-Like Modulations of Stimulus Processing during Reinforcement Learning.

PubMed

Luque, David; Beesley, Tom; Morris, Richard W; Jack, Bradley N; Griffiths, Oren; Whitford, Thomas J; Le Pelley, Mike E

2017-03-15

Recent research has shown that perceptual processing of stimuli previously associated with high-value rewards is automatically prioritized even when rewards are no longer available. It has been hypothesized that such reward-related modulation of stimulus salience is conceptually similar to an "attentional habit." Recording event-related potentials in humans during a reinforcement learning task, we show strong evidence in favor of this hypothesis. Resistance to outcome devaluation (the defining feature of a habit) was shown by the stimulus-locked P1 component, reflecting activity in the extrastriate visual cortex. Analysis at longer latencies revealed a positive component (corresponding to the P3b, from 550-700 ms) sensitive to outcome devaluation. Therefore, distinct spatiotemporal patterns of brain activity were observed corresponding to habitual and goal-directed processes. These results demonstrate that reinforcement learning engages both attentional habits and goal-directed processes in parallel. Consequences for brain and computational models of reinforcement learning are discussed. SIGNIFICANCE STATEMENT The human attentional network adapts to detect stimuli that predict important rewards. A recent hypothesis suggests that the visual cortex automatically prioritizes reward-related stimuli, driven by cached representations of reward value; that is, stimulus-response habits. Alternatively, the neural system may track the current value of the predicted outcome. Our results demonstrate for the first time that visual cortex activity is increased for reward-related stimuli even when the rewarding event is temporarily devalued. In contrast, longer-latency brain activity was specifically sensitive to transient changes in reward value. Therefore, we show that both habit-like attention and goal-directed processes occur in the same learning episode at different latencies. This result has important consequences for computational models of reinforcement learning

Identifying cognitive remediation change through computational modelling--effects on reinforcement learning in schizophrenia.

PubMed

Cella, Matteo; Bishara, Anthony J; Medin, Evelina; Swan, Sarah; Reeder, Clare; Wykes, Til

2014-11-01

Converging research suggests that individuals with schizophrenia show a marked impairment in reinforcement learning, particularly in tasks requiring flexibility and adaptation. The problem has been associated with dopamine reward systems. This study explores, for the first time, the characteristics of this impairment and how it is affected by a behavioral intervention-cognitive remediation. Using computational modelling, 3 reinforcement learning parameters based on the Wisconsin Card Sorting Test (WCST) trial-by-trial performance were estimated: R (reward sensitivity), P (punishment sensitivity), and D (choice consistency). In Study 1 the parameters were compared between a group of individuals with schizophrenia (n = 100) and a healthy control group (n = 50). In Study 2 the effect of cognitive remediation therapy (CRT) on these parameters was assessed in 2 groups of individuals with schizophrenia, one receiving CRT (n = 37) and the other receiving treatment as usual (TAU, n = 34). In Study 1 individuals with schizophrenia showed impairment in the R and P parameters compared with healthy controls. Study 2 demonstrated that sensitivity to negative feedback (P) and reward (R) improved in the CRT group after therapy compared with the TAU group. R and P parameter change correlated with WCST outputs. Improvements in R and P after CRT were associated with working memory gains and reduction of negative symptoms, respectively. Schizophrenia reinforcement learning difficulties negatively influence performance in shift learning tasks. CRT can improve sensitivity to reward and punishment. Identifying parameters that show change may be useful in experimental medicine studies to identify cognitive domains susceptible to improvement. © The Author 2013. Published by Oxford University Press on behalf of the Maryland Psychiatric Research Center. All rights reserved. For permissions, please email: journals.permissions@oup.com.

Deficient reinforcement learning in medial frontal cortex as a model of dopamine-related motivational deficits in ADHD.

PubMed

Silvetti, Massimo; Wiersema, Jan R; Sonuga-Barke, Edmund; Verguts, Tom

2013-10-01

Attention Deficit/Hyperactivity Disorder (ADHD) is a pathophysiologically complex and heterogeneous condition with both cognitive and motivational components. We propose a novel computational hypothesis of motivational deficits in ADHD, drawing together recent evidence on the role of anterior cingulate cortex (ACC) and associated mesolimbic dopamine circuits in both reinforcement learning and ADHD. Based on findings of dopamine dysregulation and ACC involvement in ADHD we simulated a lesion in a previously validated computational model of ACC (Reward Value and Prediction Model, RVPM). We explored the effects of the lesion on the processing of reinforcement signals. We tested specific behavioral predictions about the profile of reinforcement-related deficits in ADHD in three experimental contexts; probability tracking task, partial and continuous reward schedules, and immediate versus delayed rewards. In addition, predictions were made at the neurophysiological level. Behavioral and neurophysiological predictions from the RVPM-based lesion-model of motivational dysfunction in ADHD were confirmed by data from previously published studies. RVPM represents a promising model of ADHD reinforcement learning suggesting that ACC dysregulation might play a role in the pathogenesis of motivational deficits in ADHD. However, more behavioral and neurophysiological studies are required to test core predictions of the model. In addition, the interaction with different brain networks underpinning other aspects of ADHD neuropathology (i.e., executive function) needs to be better understood. Copyright © 2013 Elsevier Ltd. All rights reserved.

Multiagent cooperation and competition with deep reinforcement learning.

PubMed

Tampuu, Ardi; Matiisen, Tambet; Kodelja, Dorian; Kuzovkin, Ilya; Korjus, Kristjan; Aru, Juhan; Aru, Jaan; Vicente, Raul

2017-01-01

Evolution of cooperation and competition can appear when multiple adaptive agents share a biological, social, or technological niche. In the present work we study how cooperation and competition emerge between autonomous agents that learn by reinforcement while using only their raw visual input as the state representation. In particular, we extend the Deep Q-Learning framework to multiagent environments to investigate the interaction between two learning agents in the well-known video game Pong. By manipulating the classical rewarding scheme of Pong we show how competitive and collaborative behaviors emerge. We also describe the progression from competitive to collaborative behavior when the incentive to cooperate is increased. Finally we show how learning by playing against another adaptive agent, instead of against a hard-wired algorithm, results in more robust strategies. The present work shows that Deep Q-Networks can become a useful tool for studying decentralized learning of multiagent systems coping with high-dimensional environments.

Emotional Multiagent Reinforcement Learning in Spatial Social Dilemmas.

PubMed

Yu, Chao; Zhang, Minjie; Ren, Fenghui; Tan, Guozhen

2015-12-01

Social dilemmas have attracted extensive interest in the research of multiagent systems in order to study the emergence of cooperative behaviors among selfish agents. Understanding how agents can achieve cooperation in social dilemmas through learning from local experience is a critical problem that has motivated researchers for decades. This paper investigates the possibility of exploiting emotions in agent learning in order to facilitate the emergence of cooperation in social dilemmas. In particular, the spatial version of social dilemmas is considered to study the impact of local interactions on the emergence of cooperation in the whole system. A double-layered emotional multiagent reinforcement learning framework is proposed to endow agents with internal cognitive and emotional capabilities that can drive these agents to learn cooperative behaviors. Experimental results reveal that various network topologies and agent heterogeneities have significant impacts on agent learning behaviors in the proposed framework, and under certain circumstances, high levels of cooperation can be achieved among the agents.

Multiagent cooperation and competition with deep reinforcement learning

PubMed Central

Kodelja, Dorian; Kuzovkin, Ilya; Korjus, Kristjan; Aru, Juhan; Aru, Jaan; Vicente, Raul

2017-01-01

Evolution of cooperation and competition can appear when multiple adaptive agents share a biological, social, or technological niche. In the present work we study how cooperation and competition emerge between autonomous agents that learn by reinforcement while using only their raw visual input as the state representation. In particular, we extend the Deep Q-Learning framework to multiagent environments to investigate the interaction between two learning agents in the well-known video game Pong. By manipulating the classical rewarding scheme of Pong we show how competitive and collaborative behaviors emerge. We also describe the progression from competitive to collaborative behavior when the incentive to cooperate is increased. Finally we show how learning by playing against another adaptive agent, instead of against a hard-wired algorithm, results in more robust strategies. The present work shows that Deep Q-Networks can become a useful tool for studying decentralized learning of multiagent systems coping with high-dimensional environments. PMID:28380078

Reinforcement Learning in Young Adults with Developmental Language Impairment

ERIC Educational Resources Information Center

Lee, Joanna C.; Tomblin, J. Bruce

2012-01-01

The aim of the study was to examine reinforcement learning (RL) in young adults with developmental language impairment (DLI) within the context of a neurocomputational model of the basal ganglia-dopamine system (Frank, Seeberger, & O'Reilly, 2004). Two groups of young adults, one with DLI and the other without, were recruited. A probabilistic…

Tunnel Ventilation Control Using Reinforcement Learning Methodology

NASA Astrophysics Data System (ADS)

Chu, Baeksuk; Kim, Dongnam; Hong, Daehie; Park, Jooyoung; Chung, Jin Taek; Kim, Tae-Hyung

The main purpose of tunnel ventilation system is to maintain CO pollutant concentration and VI (visibility index) under an adequate level to provide drivers with comfortable and safe driving environment. Moreover, it is necessary to minimize power consumption used to operate ventilation system. To achieve the objectives, the control algorithm used in this research is reinforcement learning (RL) method. RL is a goal-directed learning of a mapping from situations to actions without relying on exemplary supervision or complete models of the environment. The goal of RL is to maximize a reward which is an evaluative feedback from the environment. In the process of constructing the reward of the tunnel ventilation system, two objectives listed above are included, that is, maintaining an adequate level of pollutants and minimizing power consumption. RL algorithm based on actor-critic architecture and gradient-following algorithm is adopted to the tunnel ventilation system. The simulations results performed with real data collected from existing tunnel ventilation system and real experimental verification are provided in this paper. It is confirmed that with the suggested controller, the pollutant level inside the tunnel was well maintained under allowable limit and the performance of energy consumption was improved compared to conventional control scheme.

Multiagent Reinforcement Learning With Sparse Interactions by Negotiation and Knowledge Transfer.

PubMed

Zhou, Luowei; Yang, Pei; Chen, Chunlin; Gao, Yang

2017-05-01

Reinforcement learning has significant applications for multiagent systems, especially in unknown dynamic environments. However, most multiagent reinforcement learning (MARL) algorithms suffer from such problems as exponential computation complexity in the joint state-action space, which makes it difficult to scale up to realistic multiagent problems. In this paper, a novel algorithm named negotiation-based MARL with sparse interactions (NegoSIs) is presented. In contrast to traditional sparse-interaction-based MARL algorithms, NegoSI adopts the equilibrium concept and makes it possible for agents to select the nonstrict equilibrium-dominating strategy profile (nonstrict EDSP) or meta equilibrium for their joint actions. The presented NegoSI algorithm consists of four parts: 1) the equilibrium-based framework for sparse interactions; 2) the negotiation for the equilibrium set; 3) the minimum variance method for selecting one joint action; and 4) the knowledge transfer of local Q -values. In this integrated algorithm, three techniques, i.e., unshared value functions, equilibrium solutions, and sparse interactions are adopted to achieve privacy protection, better coordination and lower computational complexity, respectively. To evaluate the performance of the presented NegoSI algorithm, two groups of experiments are carried out regarding three criteria: 1) steps of each episode; 2) rewards of each episode; and 3) average runtime. The first group of experiments is conducted using six grid world games and shows fast convergence and high scalability of the presented algorithm. Then in the second group of experiments NegoSI is applied to an intelligent warehouse problem and simulated results demonstrate the effectiveness of the presented NegoSI algorithm compared with other state-of-the-art MARL algorithms.

Reinforcement learning algorithms for robotic navigation in dynamic environments.

PubMed

Yen, Gary G; Hickey, Travis W

2004-04-01

The purpose of this study was to examine improvements to reinforcement learning (RL) algorithms in order to successfully interact within dynamic environments. The scope of the research was that of RL algorithms as applied to robotic navigation. Proposed improvements include: addition of a forgetting mechanism, use of feature based state inputs, and hierarchical structuring of an RL agent. Simulations were performed to evaluate the individual merits and flaws of each proposal, to compare proposed methods to prior established methods, and to compare proposed methods to theoretically optimal solutions. Incorporation of a forgetting mechanism did considerably improve the learning times of RL agents in a dynamic environment. However, direct implementation of a feature-based RL agent did not result in any performance enhancements, as pure feature-based navigation results in a lack of positional awareness, and the inability of the agent to determine the location of the goal state. Inclusion of a hierarchical structure in an RL agent resulted in significantly improved performance, specifically when one layer of the hierarchy included a feature-based agent for obstacle avoidance, and a standard RL agent for global navigation. In summary, the inclusion of a forgetting mechanism, and the use of a hierarchically structured RL agent offer substantially increased performance when compared to traditional RL agents navigating in a dynamic environment.

Stimulus discriminability may bias value-based probabilistic learning.

PubMed

Schutte, Iris; Slagter, Heleen A; Collins, Anne G E; Frank, Michael J; Kenemans, J Leon

2017-01-01

Reinforcement learning tasks are often used to assess participants' tendency to learn more from the positive or more from the negative consequences of one's action. However, this assessment often requires comparison in learning performance across different task conditions, which may differ in the relative salience or discriminability of the stimuli associated with more and less rewarding outcomes, respectively. To address this issue, in a first set of studies, participants were subjected to two versions of a common probabilistic learning task. The two versions differed with respect to the stimulus (Hiragana) characters associated with reward probability. The assignment of character to reward probability was fixed within version but reversed between versions. We found that performance was highly influenced by task version, which could be explained by the relative perceptual discriminability of characters assigned to high or low reward probabilities, as assessed by a separate discrimination experiment. Participants were more reliable in selecting rewarding characters that were more discriminable, leading to differences in learning curves and their sensitivity to reward probability. This difference in experienced reinforcement history was accompanied by performance biases in a test phase assessing ability to learn from positive vs. negative outcomes. In a subsequent large-scale web-based experiment, this impact of task version on learning and test measures was replicated and extended. Collectively, these findings imply a key role for perceptual factors in guiding reward learning and underscore the need to control stimulus discriminability when making inferences about individual differences in reinforcement learning.

Interactions Among Working Memory, Reinforcement Learning, and Effort in Value-Based Choice: A New Paradigm and Selective Deficits in Schizophrenia.

PubMed

Collins, Anne G E; Albrecht, Matthew A; Waltz, James A; Gold, James M; Frank, Michael J

2017-09-15

When studying learning, researchers directly observe only the participants' choices, which are often assumed to arise from a unitary learning process. However, a number of separable systems, such as working memory (WM) and reinforcement learning (RL), contribute simultaneously to human learning. Identifying each system's contributions is essential for mapping the neural substrates contributing in parallel to behavior; computational modeling can help to design tasks that allow such a separable identification of processes and infer their contributions in individuals. We present a new experimental protocol that separately identifies the contributions of RL and WM to learning, is sensitive to parametric variations in both, and allows us to investigate whether the processes interact. In experiments 1 and 2, we tested this protocol with healthy young adults (n = 29 and n = 52, respectively). In experiment 3, we used it to investigate learning deficits in medicated individuals with schizophrenia (n = 49 patients, n = 32 control subjects). Experiments 1 and 2 established WM and RL contributions to learning, as evidenced by parametric modulations of choice by load and delay and reward history, respectively. They also showed interactions between WM and RL, where RL was enhanced under high WM load. Moreover, we observed a cost of mental effort when controlling for reinforcement history: participants preferred stimuli they encountered under low WM load. Experiment 3 revealed selective deficits in WM contributions and preserved RL value learning in individuals with schizophrenia compared with control subjects. Computational approaches allow us to disentangle contributions of multiple systems to learning and, consequently, to further our understanding of psychiatric diseases. Copyright © 2017 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved.

How much of reinforcement learning is working memory, not reinforcement learning? A behavioral, computational, and neurogenetic analysis

PubMed Central

Collins, Anne G. E.; Frank, Michael J.

2012-01-01

Instrumental learning involves corticostriatal circuitry and the dopaminergic system. This system is typically modeled in the reinforcement learning (RL) framework by incrementally accumulating reward values of states and actions. However, human learning also implicates prefrontal cortical mechanisms involved in higher level cognitive functions. The interaction of these systems remains poorly understood, and models of human behavior often ignore working memory (WM) and therefore incorrectly assign behavioral variance to the RL system. Here we designed a task that highlights the profound entanglement of these two processes, even in simple learning problems. By systematically varying the size of the learning problem and delay between stimulus repetitions, we separately extracted WM-specific effects of load and delay on learning. We propose a new computational model that accounts for the dynamic integration of RL and WM processes observed in subjects' behavior. Incorporating capacity-limited WM into the model allowed us to capture behavioral variance that could not be captured in a pure RL framework even if we (implausibly) allowed separate RL systems for each set size. The WM component also allowed for a more reasonable estimation of a single RL process. Finally, we report effects of two genetic polymorphisms having relative specificity for prefrontal and basal ganglia functions. Whereas the COMT gene coding for catechol-O-methyl transferase selectively influenced model estimates of WM capacity, the GPR6 gene coding for G-protein-coupled receptor 6 influenced the RL learning rate. Thus, this study allowed us to specify distinct influences of the high-level and low-level cognitive functions on instrumental learning, beyond the possibilities offered by simple RL models. PMID:22487033

A junction-tree based learning algorithm to optimize network wide traffic control: A coordinated multi-agent framework

DOE PAGES

Zhu, Feng; Aziz, H. M. Abdul; Qian, Xinwu; ...

2015-01-31

Our study develops a novel reinforcement learning algorithm for the challenging coordinated signal control problem. Traffic signals are modeled as intelligent agents interacting with the stochastic traffic environment. The model is built on the framework of coordinated reinforcement learning. The Junction Tree Algorithm (JTA) based reinforcement learning is proposed to obtain an exact inference of the best joint actions for all the coordinated intersections. Moreover, the algorithm is implemented and tested with a network containing 18 signalized intersections in VISSIM. Finally, our results show that the JTA based algorithm outperforms independent learning (Q-learning), real-time adaptive learning, and fixed timing plansmore » in terms of average delay, number of stops, and vehicular emissions at the network level.« less

Dopamine-Dependent Reinforcement of Motor Skill Learning: Evidence from Gilles de la Tourette Syndrome

ERIC Educational Resources Information Center

Palminteri, Stefano; Lebreton, Mael; Worbe, Yulia; Hartmann, Andreas; Lehericy, Stephane; Vidailhet, Marie; Grabli, David; Pessiglione, Mathias

2011-01-01

Reinforcement learning theory has been extensively used to understand the neural underpinnings of instrumental behaviour. A central assumption surrounds dopamine signalling reward prediction errors, so as to update action values and ensure better choices in the future. However, educators may share the intuitive idea that reinforcements not only…

Processing speed enhances model-based over model-free reinforcement learning in the presence of high working memory functioning

PubMed Central

Schad, Daniel J.; Jünger, Elisabeth; Sebold, Miriam; Garbusow, Maria; Bernhardt, Nadine; Javadi, Amir-Homayoun; Zimmermann, Ulrich S.; Smolka, Michael N.; Heinz, Andreas; Rapp, Michael A.; Huys, Quentin J. M.

2014-01-01

Theories of decision-making and its neural substrates have long assumed the existence of two distinct and competing valuation systems, variously described as goal-directed vs. habitual, or, more recently and based on statistical arguments, as model-free vs. model-based reinforcement-learning. Though both have been shown to control choices, the cognitive abilities associated with these systems are under ongoing investigation. Here we examine the link to cognitive abilities, and find that individual differences in processing speed covary with a shift from model-free to model-based choice control in the presence of above-average working memory function. This suggests shared cognitive and neural processes; provides a bridge between literatures on intelligence and valuation; and may guide the development of process models of different valuation components. Furthermore, it provides a rationale for individual differences in the tendency to deploy valuation systems, which may be important for understanding the manifold neuropsychiatric diseases associated with malfunctions of valuation. PMID:25566131

A Novel Dynamic Spectrum Access Framework Based on Reinforcement Learning for Cognitive Radio Sensor Networks.

PubMed

Lin, Yun; Wang, Chao; Wang, Jiaxing; Dou, Zheng

2016-10-12

Cognitive radio sensor networks are one of the kinds of application where cognitive techniques can be adopted and have many potential applications, challenges and future research trends. According to the research surveys, dynamic spectrum access is an important and necessary technology for future cognitive sensor networks. Traditional methods of dynamic spectrum access are based on spectrum holes and they have some drawbacks, such as low accessibility and high interruptibility, which negatively affect the transmission performance of the sensor networks. To address this problem, in this paper a new initialization mechanism is proposed to establish a communication link and set up a sensor network without adopting spectrum holes to convey control information. Specifically, firstly a transmission channel model for analyzing the maximum accessible capacity for three different polices in a fading environment is discussed. Secondly, a hybrid spectrum access algorithm based on a reinforcement learning model is proposed for the power allocation problem of both the transmission channel and the control channel. Finally, extensive simulations have been conducted and simulation results show that this new algorithm provides a significant improvement in terms of the tradeoff between the control channel reliability and the efficiency of the transmission channel.

A Novel Dynamic Spectrum Access Framework Based on Reinforcement Learning for Cognitive Radio Sensor Networks

PubMed Central

Lin, Yun; Wang, Chao; Wang, Jiaxing; Dou, Zheng

2016-01-01

Cognitive radio sensor networks are one of the kinds of application where cognitive techniques can be adopted and have many potential applications, challenges and future research trends. According to the research surveys, dynamic spectrum access is an important and necessary technology for future cognitive sensor networks. Traditional methods of dynamic spectrum access are based on spectrum holes and they have some drawbacks, such as low accessibility and high interruptibility, which negatively affect the transmission performance of the sensor networks. To address this problem, in this paper a new initialization mechanism is proposed to establish a communication link and set up a sensor network without adopting spectrum holes to convey control information. Specifically, firstly a transmission channel model for analyzing the maximum accessible capacity for three different polices in a fading environment is discussed. Secondly, a hybrid spectrum access algorithm based on a reinforcement learning model is proposed for the power allocation problem of both the transmission channel and the control channel. Finally, extensive simulations have been conducted and simulation results show that this new algorithm provides a significant improvement in terms of the tradeoff between the control channel reliability and the efficiency of the transmission channel. PMID:27754316

The involvement of model-based but not model-free learning signals during observational reward learning in the absence of choice.

PubMed

Dunne, Simon; D'Souza, Arun; O'Doherty, John P

2016-06-01

A major open question is whether computational strategies thought to be used during experiential learning, specifically model-based and model-free reinforcement learning, also support observational learning. Furthermore, the question of how observational learning occurs when observers must learn about the value of options from observing outcomes in the absence of choice has not been addressed. In the present study we used a multi-armed bandit task that encouraged human participants to employ both experiential and observational learning while they underwent functional magnetic resonance imaging (fMRI). We found evidence for the presence of model-based learning signals during both observational and experiential learning in the intraparietal sulcus. However, unlike during experiential learning, model-free learning signals in the ventral striatum were not detectable during this form of observational learning. These results provide insight into the flexibility of the model-based learning system, implicating this system in learning during observation as well as from direct experience, and further suggest that the model-free reinforcement learning system may be less flexible with regard to its involvement in observational learning. Copyright © 2016 the American Physiological Society.

Gaze-contingent reinforcement learning reveals incentive value of social signals in young children and adults

PubMed Central

Smith, Tim J.; Senju, Atsushi

2017-01-01

While numerous studies have demonstrated that infants and adults preferentially orient to social stimuli, it remains unclear as to what drives such preferential orienting. It has been suggested that the learned association between social cues and subsequent reward delivery might shape such social orienting. Using a novel, spontaneous indication of reinforcement learning (with the use of a gaze contingent reward-learning task), we investigated whether children and adults' orienting towards social and non-social visual cues can be elicited by the association between participants' visual attention and a rewarding outcome. Critically, we assessed whether the engaging nature of the social cues influences the process of reinforcement learning. Both children and adults learned to orient more often to the visual cues associated with reward delivery, demonstrating that cue–reward association reinforced visual orienting. More importantly, when the reward-predictive cue was social and engaging, both children and adults learned the cue–reward association faster and more efficiently than when the reward-predictive cue was social but non-engaging. These new findings indicate that social engaging cues have a positive incentive value. This could possibly be because they usually coincide with positive outcomes in real life, which could partly drive the development of social orienting. PMID:28250186

Gaze-contingent reinforcement learning reveals incentive value of social signals in young children and adults.

PubMed

Vernetti, Angélina; Smith, Tim J; Senju, Atsushi

2017-03-15

While numerous studies have demonstrated that infants and adults preferentially orient to social stimuli, it remains unclear as to what drives such preferential orienting. It has been suggested that the learned association between social cues and subsequent reward delivery might shape such social orienting. Using a novel, spontaneous indication of reinforcement learning (with the use of a gaze contingent reward-learning task), we investigated whether children and adults' orienting towards social and non-social visual cues can be elicited by the association between participants' visual attention and a rewarding outcome. Critically, we assessed whether the engaging nature of the social cues influences the process of reinforcement learning. Both children and adults learned to orient more often to the visual cues associated with reward delivery, demonstrating that cue-reward association reinforced visual orienting. More importantly, when the reward-predictive cue was social and engaging, both children and adults learned the cue-reward association faster and more efficiently than when the reward-predictive cue was social but non-engaging. These new findings indicate that social engaging cues have a positive incentive value. This could possibly be because they usually coincide with positive outcomes in real life, which could partly drive the development of social orienting. © 2017 The Authors.

Learning alternative movement coordination patterns using reinforcement feedback.

PubMed

Lin, Tzu-Hsiang; Denomme, Amber; Ranganathan, Rajiv

2018-05-01

One of the characteristic features of the human motor system is redundancy-i.e., the ability to achieve a given task outcome using multiple coordination patterns. However, once participants settle on using a specific coordination pattern, the process of learning to use a new alternative coordination pattern to perform the same task is still poorly understood. Here, using two experiments, we examined this process of how participants shift from one coordination pattern to another using different reinforcement schedules. Participants performed a virtual reaching task, where they moved a cursor to different targets positioned on the screen. Our goal was to make participants use a coordination pattern with greater trunk motion, and to this end, we provided reinforcement by making the cursor disappear if the trunk motion during the reach did not cross a specified threshold value. In Experiment 1, we compared two reinforcement schedules in two groups of participants-an abrupt group, where the threshold was introduced immediately at the beginning of practice; and a gradual group, where the threshold was introduced gradually with practice. Results showed that both abrupt and gradual groups were effective in shifting their coordination patterns to involve greater trunk motion, but the abrupt group showed greater retention when the reinforcement was removed. In Experiment 2, we examined the basis of this advantage in the abrupt group using two additional control groups. Results showed that the advantage of the abrupt group was because of a greater number of practice trials with the desired coordination pattern. Overall, these results show that reinforcement can be successfully used to shift coordination patterns, which has potential in the rehabilitation of movement disorders.

Habits, action sequences, and reinforcement learning

PubMed Central

Dezfouli, Amir; Balleine, Bernard W.

2012-01-01

It is now widely accepted that instrumental actions can be either goal-directed or habitual; whereas the former are rapidly acquire and regulated by their outcome, the latter are reflexive, elicited by antecedent stimuli rather than their consequences. Model-based reinforcement learning (RL) provides an elegant description of goal-directed action. Through exposure to states, actions and rewards, the agent rapidly constructs a model of the world and can choose an appropriate action based on quite abstract changes in environmental and evaluative demands. This model is powerful but has a problem explaining the development of habitual actions. To account for habits, theorists have argued that another action controller is required, called model-free RL, that does not form a model of the world but rather caches action values within states allowing a state to select an action based on its reward history rather than its consequences. Nevertheless, there are persistent problems with important predictions from the model; most notably the failure of model-free RL correctly to predict the insensitivity of habitual actions to changes in the action-reward contingency. Here, we suggest that introducing model-free RL in instrumental conditioning is unnecessary and demonstrate that reconceptualizing habits as action sequences allows model-based RL to be applied to both goal-directed and habitual actions in a manner consistent with what real animals do. This approach has significant implications for the way habits are currently investigated and generates new experimental predictions. PMID:22487034

Reinforcement learning solution for HJB equation arising in constrained optimal control problem.

PubMed

Luo, Biao; Wu, Huai-Ning; Huang, Tingwen; Liu, Derong

2015-11-01

The constrained optimal control problem depends on the solution of the complicated Hamilton-Jacobi-Bellman equation (HJBE). In this paper, a data-based off-policy reinforcement learning (RL) method is proposed, which learns the solution of the HJBE and the optimal control policy from real system data. One important feature of the off-policy RL is that its policy evaluation can be realized with data generated by other behavior policies, not necessarily the target policy, which solves the insufficient exploration problem. The convergence of the off-policy RL is proved by demonstrating its equivalence to the successive approximation approach. Its implementation procedure is based on the actor-critic neural networks structure, where the function approximation is conducted with linearly independent basis functions. Subsequently, the convergence of the implementation procedure with function approximation is also proved. Finally, its effectiveness is verified through computer simulations. Copyright © 2015 Elsevier Ltd. All rights reserved.

Reusable Reinforcement Learning via Shallow Trails.

PubMed

Yu, Yang; Chen, Shi-Yong; Da, Qing; Zhou, Zhi-Hua

2018-06-01

Reinforcement learning has shown great success in helping learning agents accomplish tasks autonomously from environment interactions. Meanwhile in many real-world applications, an agent needs to accomplish not only a fixed task but also a range of tasks. For this goal, an agent can learn a metapolicy over a set of training tasks that are drawn from an underlying distribution. By maximizing the total reward summed over all the training tasks, the metapolicy can then be reused in accomplishing test tasks from the same distribution. However, in practice, we face two major obstacles to train and reuse metapolicies well. First, how to identify tasks that are unrelated or even opposite with each other, in order to avoid their mutual interference in the training. Second, how to characterize task features, according to which a metapolicy can be reused. In this paper, we propose the MetA-Policy LEarning (MAPLE) approach that overcomes the two difficulties by introducing the shallow trail. It probes a task by running a roughly trained policy. Using the rewards of the shallow trail, MAPLE automatically groups similar tasks. Moreover, when the task parameters are unknown, the rewards of the shallow trail also serve as task features. Empirical studies on several controlling tasks verify that MAPLE can train metapolicies well and receives high reward on test tasks.

Explicit and implicit reinforcement learning across the psychosis spectrum.

PubMed

Barch, Deanna M; Carter, Cameron S; Gold, James M; Johnson, Sheri L; Kring, Ann M; MacDonald, Angus W; Pizzagalli, Diego A; Ragland, J Daniel; Silverstein, Steven M; Strauss, Milton E

2017-07-01

Motivational and hedonic impairments are core features of a variety of types of psychopathology. An important aspect of motivational function is reinforcement learning (RL), including implicit (i.e., outside of conscious awareness) and explicit (i.e., including explicit representations about potential reward associations) learning, as well as both positive reinforcement (learning about actions that lead to reward) and punishment (learning to avoid actions that lead to loss). Here we present data from paradigms designed to assess both positive and negative components of both implicit and explicit RL, examine performance on each of these tasks among individuals with schizophrenia, schizoaffective disorder, and bipolar disorder with psychosis, and examine their relative relationships to specific symptom domains transdiagnostically. None of the diagnostic groups differed significantly from controls on the implicit RL tasks in either bias toward a rewarded response or bias away from a punished response. However, on the explicit RL task, both the individuals with schizophrenia and schizoaffective disorder performed significantly worse than controls, but the individuals with bipolar did not. Worse performance on the explicit RL task, but not the implicit RL task, was related to worse motivation and pleasure symptoms across all diagnostic categories. Performance on explicit RL, but not implicit RL, was related to working memory, which accounted for some of the diagnostic group differences. However, working memory did not account for the relationship of explicit RL to motivation and pleasure symptoms. These findings suggest transdiagnostic relationships across the spectrum of psychotic disorders between motivation and pleasure impairments and explicit RL. (PsycINFO Database Record (c) 2017 APA, all rights reserved).

Off-policy reinforcement learning for H∞ control design.

PubMed

Luo, Biao; Wu, Huai-Ning; Huang, Tingwen

2015-01-01

The H∞ control design problem is considered for nonlinear systems with unknown internal system model. It is known that the nonlinear H∞ control problem can be transformed into solving the so-called Hamilton-Jacobi-Isaacs (HJI) equation, which is a nonlinear partial differential equation that is generally impossible to be solved analytically. Even worse, model-based approaches cannot be used for approximately solving HJI equation, when the accurate system model is unavailable or costly to obtain in practice. To overcome these difficulties, an off-policy reinforcement leaning (RL) method is introduced to learn the solution of HJI equation from real system data instead of mathematical system model, and its convergence is proved. In the off-policy RL method, the system data can be generated with arbitrary policies rather than the evaluating policy, which is extremely important and promising for practical systems. For implementation purpose, a neural network (NN)-based actor-critic structure is employed and a least-square NN weight update algorithm is derived based on the method of weighted residuals. Finally, the developed NN-based off-policy RL method is tested on a linear F16 aircraft plant, and further applied to a rotational/translational actuator system.

Framing Reinforcement Learning from Human Reward: Reward Positivity, Temporal Discounting, Episodicity, and Performance

DTIC Science & Technology

2014-09-29

Framing Reinforcement Learning from Human Reward: Reward Positivity, Temporal Discounting, Episodicity , and Performance W. Bradley Knox...positive a trainer’s reward values are; temporal discounting, the extent to which future reward is discounted in value; episodicity , whether task...learning occurs in discrete learning episodes instead of one continuing session; and task performance, the agent’s performance on the task the trainer

Optimizing Chemical Reactions with Deep Reinforcement Learning.

PubMed

Zhou, Zhenpeng; Li, Xiaocheng; Zare, Richard N

2017-12-27

Deep reinforcement learning was employed to optimize chemical reactions. Our model iteratively records the results of a chemical reaction and chooses new experimental conditions to improve the reaction outcome. This model outperformed a state-of-the-art blackbox optimization algorithm by using 71% fewer steps on both simulations and real reactions. Furthermore, we introduced an efficient exploration strategy by drawing the reaction conditions from certain probability distributions, which resulted in an improvement on regret from 0.062 to 0.039 compared with a deterministic policy. Combining the efficient exploration policy with accelerated microdroplet reactions, optimal reaction conditions were determined in 30 min for the four reactions considered, and a better understanding of the factors that control microdroplet reactions was reached. Moreover, our model showed a better performance after training on reactions with similar or even dissimilar underlying mechanisms, which demonstrates its learning ability.

A self-learning rule base for command following in dynamical systems

NASA Technical Reports Server (NTRS)

Tsai, Wei K.; Lee, Hon-Mun; Parlos, Alexander

1992-01-01

In this paper, a self-learning Rule Base for command following in dynamical systems is presented. The learning is accomplished though reinforcement learning using an associative memory called SAM. The main advantage of SAM is that it is a function approximator with explicit storage of training samples. A learning algorithm patterned after the dynamic programming is proposed. Two artificially created, unstable dynamical systems are used for testing, and the Rule Base was used to generate a feedback control to improve the command following ability of the otherwise uncontrolled systems. The numerical results are very encouraging. The controlled systems exhibit a more stable behavior and a better capability to follow reference commands. The rules resulting from the reinforcement learning are explicitly stored and they can be modified or augmented by human experts. Due to overlapping storage scheme of SAM, the stored rules are similar to fuzzy rules.

Reinforcement Learning Deficits in People with Schizophrenia Persist after Extended Trials

PubMed Central

Cicero, David C.; Martin, Elizabeth A.; Becker, Theresa M.; Kerns, John G.

2014-01-01

Previous research suggests that people with schizophrenia have difficulty learning from positive feedback and when learning needs to occur rapidly. However, they seem to have relatively intact learning from negative feedback when learning occurs gradually. Participants are typically given a limited amount of acquisition trials to learn the reward contingencies and then tested about what they learned. The current study examined whether participants with schizophrenia continue to display these deficits when given extra time to learn the contingences. Participants with schizophrenia and matched healthy controls completed the Probabilistic Selection Task, which measures positive and negative feedback learning separately. Participants with schizophrenia showed a deficit in learning from both positive and negative feedback. These reward learning deficits persisted even if people with schizophrenia are given extra time (up to 10 blocks of 60 trials) to learn the reward contingencies. These results suggest that the observed deficits cannot be attributed solely to slower learning and instead reflect a specific deficit in reinforcement learning. PMID:25172610

Intrinsic interactive reinforcement learning - Using error-related potentials for real world human-robot interaction.

PubMed

Kim, Su Kyoung; Kirchner, Elsa Andrea; Stefes, Arne; Kirchner, Frank

2017-12-14

Reinforcement learning (RL) enables robots to learn its optimal behavioral strategy in dynamic environments based on feedback. Explicit human feedback during robot RL is advantageous, since an explicit reward function can be easily adapted. However, it is very demanding and tiresome for a human to continuously and explicitly generate feedback. Therefore, the development of implicit approaches is of high relevance. In this paper, we used an error-related potential (ErrP), an event-related activity in the human electroencephalogram (EEG), as an intrinsically generated implicit feedback (rewards) for RL. Initially we validated our approach with seven subjects in a simulated robot learning scenario. ErrPs were detected online in single trial with a balanced accuracy (bACC) of 91%, which was sufficient to learn to recognize gestures and the correct mapping between human gestures and robot actions in parallel. Finally, we validated our approach in a real robot scenario, in which seven subjects freely chose gestures and the real robot correctly learned the mapping between gestures and actions (ErrP detection (90% bACC)). In this paper, we demonstrated that intrinsically generated EEG-based human feedback in RL can successfully be used to implicitly improve gesture-based robot control during human-robot interaction. We call our approach intrinsic interactive RL.

Risk-sensitive reinforcement learning.

PubMed

Shen, Yun; Tobia, Michael J; Sommer, Tobias; Obermayer, Klaus

2014-07-01

We derive a family of risk-sensitive reinforcement learning methods for agents, who face sequential decision-making tasks in uncertain environments. By applying a utility function to the temporal difference (TD) error, nonlinear transformations are effectively applied not only to the received rewards but also to the true transition probabilities of the underlying Markov decision process. When appropriate utility functions are chosen, the agents' behaviors express key features of human behavior as predicted by prospect theory (Kahneman & Tversky, 1979 ), for example, different risk preferences for gains and losses, as well as the shape of subjective probability curves. We derive a risk-sensitive Q-learning algorithm, which is necessary for modeling human behavior when transition probabilities are unknown, and prove its convergence. As a proof of principle for the applicability of the new framework, we apply it to quantify human behavior in a sequential investment task. We find that the risk-sensitive variant provides a significantly better fit to the behavioral data and that it leads to an interpretation of the subject's responses that is indeed consistent with prospect theory. The analysis of simultaneously measured fMRI signals shows a significant correlation of the risk-sensitive TD error with BOLD signal change in the ventral striatum. In addition we find a significant correlation of the risk-sensitive Q-values with neural activity in the striatum, cingulate cortex, and insula that is not present if standard Q-values are used.

Decentralized reinforcement-learning control and emergence of motion patterns

NASA Astrophysics Data System (ADS)

Svinin, Mikhail; Yamada, Kazuyaki; Okhura, Kazuhiro; Ueda, Kanji

1998-10-01

In this paper we propose a system for studying emergence of motion patterns in autonomous mobile robotic systems. The system implements an instance-based reinforcement learning control. Three spaces are of importance in formulation of the control scheme. They are the work space, the sensor space, and the action space. Important feature of our system is that all these spaces are assumed to be continuous. The core part of the system is a classifier system. Based on the sensory state space analysis, the control is decentralized and is specified at the lowest level of the control system. However, the local controllers are implicitly connected through the perceived environment information. Therefore, they constitute a dynamic environment with respect to each other. The proposed control scheme is tested under simulation for a mobile robot in a navigation task. It is shown that some patterns of global behavior--such as collision avoidance, wall-following, light-seeking--can emerge from the local controllers.

Investigation of a reinforcement-based toilet training procedure for children with autism.

PubMed

Cicero, Frank R; Pfadt, Al

2002-01-01

Independent toileting is an important developmental skill which individuals with developmental disabilities often find a challenge to master. Effective toilet training interventions have been designed which rely on a combination of basic operant principles of positive reinforcement and punishment. In the present study, the effectiveness of a reinforcement-based toilet training intervention was investigated with three children with a diagnosis of autism. Procedures included a combination of positive reinforcement, graduated guidance, scheduled practice trials and forward prompting. Results indicated that all procedures were implemented in response to urination accidents. A three participants reduced urination accidents to zero and learned to spontaneously request use of the bathroom within 7-11 days of training. Gains were maintained over 6-month and 1-year follow-ups. Findings suggest that the proposed procedure is an effective and rapid method of toilet training, which can be implemented within a structured school setting with generalization to the home environment.

FMRQ-A Multiagent Reinforcement Learning Algorithm for Fully Cooperative Tasks.

PubMed

Zhang, Zhen; Zhao, Dongbin; Gao, Junwei; Wang, Dongqing; Dai, Yujie

2017-06-01

In this paper, we propose a multiagent reinforcement learning algorithm dealing with fully cooperative tasks. The algorithm is called frequency of the maximum reward Q-learning (FMRQ). FMRQ aims to achieve one of the optimal Nash equilibria so as to optimize the performance index in multiagent systems. The frequency of obtaining the highest global immediate reward instead of immediate reward is used as the reinforcement signal. With FMRQ each agent does not need the observation of the other agents' actions and only shares its state and reward at each step. We validate FMRQ through case studies of repeated games: four cases of two-player two-action and one case of three-player two-action. It is demonstrated that FMRQ can converge to one of the optimal Nash equilibria in these cases. Moreover, comparison experiments on tasks with multiple states and finite steps are conducted. One is box-pushing and the other one is distributed sensor network problem. Experimental results show that the proposed algorithm outperforms others with higher performance.

A cholinergic feedback circuit to regulate striatal population uncertainty and optimize reinforcement learning.

PubMed

Franklin, Nicholas T; Frank, Michael J

2015-12-25

Convergent evidence suggests that the basal ganglia support reinforcement learning by adjusting action values according to reward prediction errors. However, adaptive behavior in stochastic environments requires the consideration of uncertainty to dynamically adjust the learning rate. We consider how cholinergic tonically active interneurons (TANs) may endow the striatum with such a mechanism in computational models spanning three Marr's levels of analysis. In the neural model, TANs modulate the excitability of spiny neurons, their population response to reinforcement, and hence the effective learning rate. Long TAN pauses facilitated robustness to spurious outcomes by increasing divergence in synaptic weights between neurons coding for alternative action values, whereas short TAN pauses facilitated stochastic behavior but increased responsiveness to change-points in outcome contingencies. A feedback control system allowed TAN pauses to be dynamically modulated by uncertainty across the spiny neuron population, allowing the system to self-tune and optimize performance across stochastic environments.

Developing PFC representations using reinforcement learning

PubMed Central

Reynolds, Jeremy R.; O'Reilly, Randall C.

2009-01-01

From both functional and biological considerations, it is widely believed that action production, planning, and goal-oriented behaviors supported by the frontal cortex are organized hierarchically (Fuster, 1990, Koechlin, Ody, & Kouneiher, 2003, & Miller, Galanter, & Pribram, 1960) However, the nature of the different levels of the hierarchy remains unclear, and little attention has been paid to the origins of such a hierarchy. We address these issues through biologically-inspired computational models that develop representations through reinforcement learning. We explore several different factors in these models that might plausibly give rise to a hierarchical organization of representations within the PFC, including an initial connectivity hierarchy within PFC, a hierarchical set of connections between PFC and subcortical structures controlling it, and differential synaptic plasticity schedules. Simulation results indicate that architectural constraints contribute to the segregation of different types of representations, and that this segregation facilitates learning. These findings are consistent with the idea that there is a functional hierarchy in PFC, as captured in our earlier computational models of PFC function and a growing body of empirical data. PMID:19591977

Fault-tolerant optimised tracking control for unknown discrete-time linear systems using a combined reinforcement learning and residual compensation methodology

NASA Astrophysics Data System (ADS)

Han, Ke-Zhen; Feng, Jian; Cui, Xiaohong

2017-10-01

This paper considers the fault-tolerant optimised tracking control (FTOTC) problem for unknown discrete-time linear system. A research scheme is proposed on the basis of data-based parity space identification, reinforcement learning and residual compensation techniques. The main characteristic of this research scheme lies in the parity-space-identification-based simultaneous tracking control and residual compensation. The specific technical line consists of four main contents: apply subspace aided method to design observer-based residual generator; use reinforcement Q-learning approach to solve optimised tracking control policy; rely on robust H∞ theory to achieve noise attenuation; adopt fault estimation triggered by residual generator to perform fault compensation. To clarify the design and implementation procedures, an integrated algorithm is further constructed to link up these four functional units. The detailed analysis and proof are subsequently given to explain the guaranteed FTOTC performance of the proposed conclusions. Finally, a case simulation is provided to verify its effectiveness.

Forgetting in Reinforcement Learning Links Sustained Dopamine Signals to Motivation

PubMed Central

Morita, Kenji

2016-01-01

It has been suggested that dopamine (DA) represents reward-prediction-error (RPE) defined in reinforcement learning and therefore DA responds to unpredicted but not predicted reward. However, recent studies have found DA response sustained towards predictable reward in tasks involving self-paced behavior, and suggested that this response represents a motivational signal. We have previously shown that RPE can sustain if there is decay/forgetting of learned-values, which can be implemented as decay of synaptic strengths storing learned-values. This account, however, did not explain the suggested link between tonic/sustained DA and motivation. In the present work, we explored the motivational effects of the value-decay in self-paced approach behavior, modeled as a series of ‘Go’ or ‘No-Go’ selections towards a goal. Through simulations, we found that the value-decay can enhance motivation, specifically, facilitate fast goal-reaching, albeit counterintuitively. Mathematical analyses revealed that underlying potential mechanisms are twofold: (1) decay-induced sustained RPE creates a gradient of ‘Go’ values towards a goal, and (2) value-contrasts between ‘Go’ and ‘No-Go’ are generated because while chosen values are continually updated, unchosen values simply decay. Our model provides potential explanations for the key experimental findings that suggest DA's roles in motivation: (i) slowdown of behavior by post-training blockade of DA signaling, (ii) observations that DA blockade severely impairs effortful actions to obtain rewards while largely sparing seeking of easily obtainable rewards, and (iii) relationships between the reward amount, the level of motivation reflected in the speed of behavior, and the average level of DA. These results indicate that reinforcement learning with value-decay, or forgetting, provides a parsimonious mechanistic account for the DA's roles in value-learning and motivation. Our results also suggest that when biological

Forgetting in Reinforcement Learning Links Sustained Dopamine Signals to Motivation.

PubMed

Kato, Ayaka; Morita, Kenji

2016-10-01

It has been suggested that dopamine (DA) represents reward-prediction-error (RPE) defined in reinforcement learning and therefore DA responds to unpredicted but not predicted reward. However, recent studies have found DA response sustained towards predictable reward in tasks involving self-paced behavior, and suggested that this response represents a motivational signal. We have previously shown that RPE can sustain if there is decay/forgetting of learned-values, which can be implemented as decay of synaptic strengths storing learned-values. This account, however, did not explain the suggested link between tonic/sustained DA and motivation. In the present work, we explored the motivational effects of the value-decay in self-paced approach behavior, modeled as a series of 'Go' or 'No-Go' selections towards a goal. Through simulations, we found that the value-decay can enhance motivation, specifically, facilitate fast goal-reaching, albeit counterintuitively. Mathematical analyses revealed that underlying potential mechanisms are twofold: (1) decay-induced sustained RPE creates a gradient of 'Go' values towards a goal, and (2) value-contrasts between 'Go' and 'No-Go' are generated because while chosen values are continually updated, unchosen values simply decay. Our model provides potential explanations for the key experimental findings that suggest DA's roles in motivation: (i) slowdown of behavior by post-training blockade of DA signaling, (ii) observations that DA blockade severely impairs effortful actions to obtain rewards while largely sparing seeking of easily obtainable rewards, and (iii) relationships between the reward amount, the level of motivation reflected in the speed of behavior, and the average level of DA. These results indicate that reinforcement learning with value-decay, or forgetting, provides a parsimonious mechanistic account for the DA's roles in value-learning and motivation. Our results also suggest that when biological systems for value-learning

Optimizing Chemical Reactions with Deep Reinforcement Learning

PubMed Central

2017-01-01

Deep reinforcement learning was employed to optimize chemical reactions. Our model iteratively records the results of a chemical reaction and chooses new experimental conditions to improve the reaction outcome. This model outperformed a state-of-the-art blackbox optimization algorithm by using 71% fewer steps on both simulations and real reactions. Furthermore, we introduced an efficient exploration strategy by drawing the reaction conditions from certain probability distributions, which resulted in an improvement on regret from 0.062 to 0.039 compared with a deterministic policy. Combining the efficient exploration policy with accelerated microdroplet reactions, optimal reaction conditions were determined in 30 min for the four reactions considered, and a better understanding of the factors that control microdroplet reactions was reached. Moreover, our model showed a better performance after training on reactions with similar or even dissimilar underlying mechanisms, which demonstrates its learning ability. PMID:29296675

Reinforcement-Learning-Based Robust Controller Design for Continuous-Time Uncertain Nonlinear Systems Subject to Input Constraints.

PubMed

Liu, Derong; Yang, Xiong; Wang, Ding; Wei, Qinglai

2015-07-01

The design of stabilizing controller for uncertain nonlinear systems with control constraints is a challenging problem. The constrained-input coupled with the inability to identify accurately the uncertainties motivates the design of stabilizing controller based on reinforcement-learning (RL) methods. In this paper, a novel RL-based robust adaptive control algorithm is developed for a class of continuous-time uncertain nonlinear systems subject to input constraints. The robust control problem is converted to the constrained optimal control problem with appropriately selecting value functions for the nominal system. Distinct from typical action-critic dual networks employed in RL, only one critic neural network (NN) is constructed to derive the approximate optimal control. Meanwhile, unlike initial stabilizing control often indispensable in RL, there is no special requirement imposed on the initial control. By utilizing Lyapunov's direct method, the closed-loop optimal control system and the estimated weights of the critic NN are proved to be uniformly ultimately bounded. In addition, the derived approximate optimal control is verified to guarantee the uncertain nonlinear system to be stable in the sense of uniform ultimate boundedness. Two simulation examples are provided to illustrate the effectiveness and applicability of the present approach.

A spiking neural network model of model-free reinforcement learning with high-dimensional sensory input and perceptual ambiguity.

PubMed

Nakano, Takashi; Otsuka, Makoto; Yoshimoto, Junichiro; Doya, Kenji

2015-01-01

A theoretical framework of reinforcement learning plays an important role in understanding action selection in animals. Spiking neural networks provide a theoretically grounded means to test computational hypotheses on neurally plausible algorithms of reinforcement learning through numerical simulation. However, most of these models cannot handle observations which are noisy, or occurred in the past, even though these are inevitable and constraining features of learning in real environments. This class of problem is formally known as partially observable reinforcement learning (PORL) problems. It provides a generalization of reinforcement learning to partially observable domains. In addition, observations in the real world tend to be rich and high-dimensional. In this work, we use a spiking neural network model to approximate the free energy of a restricted Boltzmann machine and apply it to the solution of PORL problems with high-dimensional observations. Our spiking network model solves maze tasks with perceptually ambiguous high-dimensional observations without knowledge of the true environment. An extended model with working memory also solves history-dependent tasks. The way spiking neural networks handle PORL problems may provide a glimpse into the underlying laws of neural information processing which can only be discovered through such a top-down approach.

A Spiking Neural Network Model of Model-Free Reinforcement Learning with High-Dimensional Sensory Input and Perceptual Ambiguity

PubMed Central

Nakano, Takashi; Otsuka, Makoto; Yoshimoto, Junichiro; Doya, Kenji

2015-01-01

A theoretical framework of reinforcement learning plays an important role in understanding action selection in animals. Spiking neural networks provide a theoretically grounded means to test computational hypotheses on neurally plausible algorithms of reinforcement learning through numerical simulation. However, most of these models cannot handle observations which are noisy, or occurred in the past, even though these are inevitable and constraining features of learning in real environments. This class of problem is formally known as partially observable reinforcement learning (PORL) problems. It provides a generalization of reinforcement learning to partially observable domains. In addition, observations in the real world tend to be rich and high-dimensional. In this work, we use a spiking neural network model to approximate the free energy of a restricted Boltzmann machine and apply it to the solution of PORL problems with high-dimensional observations. Our spiking network model solves maze tasks with perceptually ambiguous high-dimensional observations without knowledge of the true environment. An extended model with working memory also solves history-dependent tasks. The way spiking neural networks handle PORL problems may provide a glimpse into the underlying laws of neural information processing which can only be discovered through such a top-down approach. PMID:25734662

Seizure Control in a Computational Model Using a Reinforcement Learning Stimulation Paradigm.

PubMed

Nagaraj, Vivek; Lamperski, Andrew; Netoff, Theoden I

2017-11-01

Neuromodulation technologies such as vagus nerve stimulation and deep brain stimulation, have shown some efficacy in controlling seizures in medically intractable patients. However, inherent patient-to-patient variability of seizure disorders leads to a wide range of therapeutic efficacy. A patient specific approach to determining stimulation parameters may lead to increased therapeutic efficacy while minimizing stimulation energy and side effects. This paper presents a reinforcement learning algorithm that optimizes stimulation frequency for controlling seizures with minimum stimulation energy. We apply our method to a computational model called the epileptor. The epileptor model simulates inter-ictal and ictal local field potential data. In order to apply reinforcement learning to the Epileptor, we introduce a specialized reward function and state-space discretization. With the reward function and discretization fixed, we test the effectiveness of the temporal difference reinforcement learning algorithm (TD(0)). For periodic pulsatile stimulation, we derive a relation that describes, for any stimulation frequency, the minimal pulse amplitude required to suppress seizures. The TD(0) algorithm is able to identify parameters that control seizures quickly. Additionally, our results show that the TD(0) algorithm refines the stimulation frequency to minimize stimulation energy thereby converging to optimal parameters reliably. An advantage of the TD(0) algorithm is that it is adaptive so that the parameters necessary to control the seizures can change over time. We show that the algorithm can converge on the optimal solution in simulation with slow and fast inter-seizure intervals.

Amygdala and Ventral Striatum Make Distinct Contributions to Reinforcement Learning.

PubMed

Costa, Vincent D; Dal Monte, Olga; Lucas, Daniel R; Murray, Elisabeth A; Averbeck, Bruno B

2016-10-19

Reinforcement learning (RL) theories posit that dopaminergic signals are integrated within the striatum to associate choices with outcomes. Often overlooked is that the amygdala also receives dopaminergic input and is involved in Pavlovian processes that influence choice behavior. To determine the relative contributions of the ventral striatum (VS) and amygdala to appetitive RL, we tested rhesus macaques with VS or amygdala lesions on deterministic and stochastic versions of a two-arm bandit reversal learning task. When learning was characterized with an RL model relative to controls, amygdala lesions caused general decreases in learning from positive feedback and choice consistency. By comparison, VS lesions only affected learning in the stochastic task. Moreover, the VS lesions hastened the monkeys' choice reaction times, which emphasized a speed-accuracy trade-off that accounted for errors in deterministic learning. These results update standard accounts of RL by emphasizing distinct contributions of the amygdala and VS to RL. Published by Elsevier Inc.

Amygdala and ventral striatum make distinct contributions to reinforcement learning

PubMed Central

Costa, Vincent D.; Monte, Olga Dal; Lucas, Daniel R.; Murray, Elisabeth A.; Averbeck, Bruno B.

2016-01-01

Summary Reinforcement learning (RL) theories posit that dopaminergic signals are integrated within the striatum to associate choices with outcomes. Often overlooked is that the amygdala also receives dopaminergic input and is involved in Pavlovian processes that influence choice behavior. To determine the relative contributions of the ventral striatum (VS) and amygdala to appetitive RL we tested rhesus macaques with VS or amygdala lesions on deterministic and stochastic versions of a two-arm bandit reversal learning task. When learning was characterized with a RL model relative to controls, amygdala lesions caused general decreases in learning from positive feedback and choice consistency. By comparison, VS lesions only affected learning in the stochastic task. Moreover, the VS lesions hastened the monkeys’ choice reaction times, which emphasized a speed-accuracy tradeoff that accounted for errors in deterministic learning. These results update standard accounts of RL by emphasizing distinct contributions of the amygdala and VS to RL. PMID:27720488

Active-learning strategies: the use of a game to reinforce learning in nursing education. A case study.

PubMed

Boctor, Lisa

2013-03-01

The majority of nursing students are kinesthetic learners, preferring a hands-on, active approach to education. Research shows that active-learning strategies can increase student learning and satisfaction. This study looks at the use of one active-learning strategy, a Jeopardy-style game, 'Nursopardy', to reinforce Fundamentals of Nursing material, aiding in students' preparation for a standardized final exam. The game was created keeping students varied learning styles and the NCLEX blueprint in mind. The blueprint was used to create 5 categories, with 26 total questions. Student survey results, using a five-point Likert scale showed that they did find this learning method enjoyable and beneficial to learning. More research is recommended regarding learning outcomes, when using active-learning strategies, such as games. Copyright © 2012 Elsevier Ltd. All rights reserved.

Deep Direct Reinforcement Learning for Financial Signal Representation and Trading.

PubMed

Deng, Yue; Bao, Feng; Kong, Youyong; Ren, Zhiquan; Dai, Qionghai

2017-03-01

Can we train the computer to beat experienced traders for financial assert trading? In this paper, we try to address this challenge by introducing a recurrent deep neural network (NN) for real-time financial signal representation and trading. Our model is inspired by two biological-related learning concepts of deep learning (DL) and reinforcement learning (RL). In the framework, the DL part automatically senses the dynamic market condition for informative feature learning. Then, the RL module interacts with deep representations and makes trading decisions to accumulate the ultimate rewards in an unknown environment. The learning system is implemented in a complex NN that exhibits both the deep and recurrent structures. Hence, we propose a task-aware backpropagation through time method to cope with the gradient vanishing issue in deep training. The robustness of the neural system is verified on both the stock and the commodity future markets under broad testing conditions.

A cholinergic feedback circuit to regulate striatal population uncertainty and optimize reinforcement learning

PubMed Central

Franklin, Nicholas T; Frank, Michael J

2015-01-01

Convergent evidence suggests that the basal ganglia support reinforcement learning by adjusting action values according to reward prediction errors. However, adaptive behavior in stochastic environments requires the consideration of uncertainty to dynamically adjust the learning rate. We consider how cholinergic tonically active interneurons (TANs) may endow the striatum with such a mechanism in computational models spanning three Marr's levels of analysis. In the neural model, TANs modulate the excitability of spiny neurons, their population response to reinforcement, and hence the effective learning rate. Long TAN pauses facilitated robustness to spurious outcomes by increasing divergence in synaptic weights between neurons coding for alternative action values, whereas short TAN pauses facilitated stochastic behavior but increased responsiveness to change-points in outcome contingencies. A feedback control system allowed TAN pauses to be dynamically modulated by uncertainty across the spiny neuron population, allowing the system to self-tune and optimize performance across stochastic environments. DOI: http://dx.doi.org/10.7554/eLife.12029.001 PMID:26705698

Statistical Mechanics of the Delayed Reward-Based Learning with Node Perturbation

NASA Astrophysics Data System (ADS)

Hiroshi Saito,; Kentaro Katahira,; Kazuo Okanoya,; Masato Okada,

2010-06-01

In reward-based learning, reward is typically given with some delay after a behavior that causes the reward. In machine learning literature, the framework of the eligibility trace has been used as one of the solutions to handle the delayed reward in reinforcement learning. In recent studies, the eligibility trace is implied to be important for difficult neuroscience problem known as the “distal reward problem”. Node perturbation is one of the stochastic gradient methods from among many kinds of reinforcement learning implementations, and it searches the approximate gradient by introducing perturbation to a network. Since the stochastic gradient method does not require a objective function differential, it is expected to be able to account for the learning mechanism of a complex system, like a brain. We study the node perturbation with the eligibility trace as a specific example of delayed reward-based learning, and analyzed it using a statistical mechanics approach. As a result, we show the optimal time constant of the eligibility trace respect to the reward delay and the existence of unlearnable parameter configurations.

Optimal Reward Functions in Distributed Reinforcement Learning

NASA Technical Reports Server (NTRS)

Wolpert, David H.; Tumer, Kagan

2000-01-01

We consider the design of multi-agent systems so as to optimize an overall world utility function when (1) those systems lack centralized communication and control, and (2) each agents runs a distinct Reinforcement Learning (RL) algorithm. A crucial issue in such design problems is to initialize/update each agent's private utility function, so as to induce best possible world utility. Traditional 'team game' solutions to this problem sidestep this issue and simply assign to each agent the world utility as its private utility function. In previous work we used the 'Collective Intelligence' framework to derive a better choice of private utility functions, one that results in world utility performance up to orders of magnitude superior to that ensuing from use of the team game utility. In this paper we extend these results. We derive the general class of private utility functions that both are easy for the individual agents to learn and that, if learned well, result in high world utility. We demonstrate experimentally that using these new utility functions can result in significantly improved performance over that of our previously proposed utility, over and above that previous utility's superiority to the conventional team game utility.

Reinforcement learning deficits in people with schizophrenia persist after extended trials.

PubMed

Cicero, David C; Martin, Elizabeth A; Becker, Theresa M; Kerns, John G

2014-12-30

Previous research suggests that people with schizophrenia have difficulty learning from positive feedback and when learning needs to occur rapidly. However, they seem to have relatively intact learning from negative feedback when learning occurs gradually. Participants are typically given a limited amount of acquisition trials to learn the reward contingencies and then tested about what they learned. The current study examined whether participants with schizophrenia continue to display these deficits when given extra time to learn the contingences. Participants with schizophrenia and matched healthy controls completed the Probabilistic Selection Task, which measures positive and negative feedback learning separately. Participants with schizophrenia showed a deficit in learning from both positive feedback and negative feedback. These reward learning deficits persisted even if people with schizophrenia are given extra time (up to 10 blocks of 60 trials) to learn the reward contingencies. These results suggest that the observed deficits cannot be attributed solely to slower learning and instead reflect a specific deficit in reinforcement learning. Copyright © 2014 Elsevier Ireland Ltd. All rights reserved.

Development and Evaluation of Mechatronics Learning System in a Web-Based Environment

ERIC Educational Resources Information Center

Shyr, Wen-Jye

2011-01-01

The development of remote laboratory suitable for the reinforcement of undergraduate level teaching of mechatronics is important. For the reason, a Web-based mechatronics learning system, called the RECOLAB (REmote COntrol LABoratory), for remote learning in engineering education has been developed in this study. The web-based environment is an…

High and low temperatures have unequal reinforcing properties in Drosophila spatial learning.

PubMed

Zars, Melissa; Zars, Troy

2006-07-01

Small insects regulate their body temperature solely through behavior. Thus, sensing environmental temperature and implementing an appropriate behavioral strategy can be critical for survival. The fly Drosophila melanogaster prefers 24 degrees C, avoiding higher and lower temperatures when tested on a temperature gradient. Furthermore, temperatures above 24 degrees C have negative reinforcing properties. In contrast, we found that flies have a preference in operant learning experiments for a low-temperature-associated position rather than the 24 degrees C alternative in the heat-box. Two additional differences between high- and low-temperature reinforcement, i.e., temperatures above and below 24 degrees C, were found. Temperatures equally above and below 24 degrees C did not reinforce equally and only high temperatures supported increased memory performance with reversal conditioning. Finally, low- and high-temperature reinforced memories are similarly sensitive to two genetic mutations. Together these results indicate the qualitative meaning of temperatures below 24 degrees C depends on the dynamics of the temperatures encountered and that the reinforcing effects of these temperatures depend on at least some common genetic components. Conceptualizing these results using the Wolf-Heisenberg model of operant conditioning, we propose the maximum difference in experienced temperatures determines the magnitude of the reinforcement input to a conditioning circuit.

Knockout crickets for the study of learning and memory: Dopamine receptor Dop1 mediates aversive but not appetitive reinforcement in crickets.

PubMed

Awata, Hiroko; Watanabe, Takahito; Hamanaka, Yoshitaka; Mito, Taro; Noji, Sumihare; Mizunami, Makoto

2015-11-02

Elucidation of reinforcement mechanisms in associative learning is an important subject in neuroscience. In mammals, dopamine neurons are thought to play critical roles in mediating both appetitive and aversive reinforcement. Our pharmacological studies suggested that octopamine and dopamine neurons mediate reward and punishment, respectively, in crickets, but recent studies in fruit-flies concluded that dopamine neurons mediates both reward and punishment, via the type 1 dopamine receptor Dop1. To resolve the discrepancy between studies in different insect species, we produced Dop1 knockout crickets using the CRISPR/Cas9 system and found that they are defective in aversive learning with sodium chloride punishment but not appetitive learning with water or sucrose reward. The results suggest that dopamine and octopamine neurons mediate aversive and appetitive reinforcement, respectively, in crickets. We suggest unexpected diversity in neurotransmitters mediating appetitive reinforcement between crickets and fruit-flies, although the neurotransmitter mediating aversive reinforcement is conserved. This study demonstrates usefulness of the CRISPR/Cas9 system for producing knockout animals for the study of learning and memory.

Assist-as-needed robotic trainer based on reinforcement learning and its application to dart-throwing.

PubMed

Obayashi, Chihiro; Tamei, Tomoya; Shibata, Tomohiro

2014-05-01

This paper proposes a novel robotic trainer for motor skill learning. It is user-adaptive inspired by the assist-as-needed principle well known in the field of physical therapy. Most previous studies in the field of the robotic assistance of motor skill learning have used predetermined desired trajectories, and it has not been examined intensively whether these trajectories were optimal for each user. Furthermore, the guidance hypothesis states that humans tend to rely too much on external assistive feedback, resulting in interference with the internal feedback necessary for motor skill learning. A few studies have proposed a system that adjusts its assistive strength according to the user's performance in order to prevent the user from relying too much on the robotic assistance. There are, however, problems in these studies, in that a physical model of the user's motor system is required, which is inherently difficult to construct. In this paper, we propose a framework for a robotic trainer that is user-adaptive and that neither requires a specific desired trajectory nor a physical model of the user's motor system, and we achieve this using model-free reinforcement learning. We chose dart-throwing as an example motor-learning task as it is one of the simplest throwing tasks, and its performance can easily be and quantitatively measured. Training experiments with novices, aiming at maximizing the score with the darts and minimizing the physical robotic assistance, demonstrate the feasibility and plausibility of the proposed framework. Copyright © 2014 Elsevier Ltd. All rights reserved.

A comparison of differential reinforcement procedures with children with autism.

PubMed

Boudreau, Brittany A; Vladescu, Jason C; Kodak, Tiffany M; Argott, Paul J; Kisamore, April N

2015-12-01

The current evaluation compared the effects of 2 differential reinforcement arrangements and a nondifferential reinforcement arrangement on the acquisition of tacts for 3 children with autism. Participants learned in all reinforcement-based conditions, and we discuss areas for future research in light of these findings and potential limitations. © Society for the Experimental Analysis of Behavior.

A Biologically Plausible Architecture of the Striatum to Solve Context-Dependent Reinforcement Learning Tasks.

PubMed

Shivkumar, Sabyasachi; Muralidharan, Vignesh; Chakravarthy, V Srinivasa

2017-01-01

Basal ganglia circuit is an important subcortical system of the brain thought to be responsible for reward-based learning. Striatum, the largest nucleus of the basal ganglia, serves as an input port that maps cortical information. Microanatomical studies show that the striatum is a mosaic of specialized input-output structures called striosomes and regions of the surrounding matrix called the matrisomes. We have developed a computational model of the striatum using layered self-organizing maps to capture the center-surround structure seen experimentally and explain its functional significance. We believe that these structural components could build representations of state and action spaces in different environments. The striatum model is then integrated with other components of basal ganglia, making it capable of solving reinforcement learning tasks. We have proposed a biologically plausible mechanism of action-based learning where the striosome biases the matrisome activity toward a preferred action. Several studies indicate that the striatum is critical in solving context dependent problems. We build on this hypothesis and the proposed model exploits the modularity of the striatum to efficiently solve such tasks.

A Biologically Plausible Architecture of the Striatum to Solve Context-Dependent Reinforcement Learning Tasks

PubMed Central

Shivkumar, Sabyasachi; Muralidharan, Vignesh; Chakravarthy, V. Srinivasa

2017-01-01

Basal ganglia circuit is an important subcortical system of the brain thought to be responsible for reward-based learning. Striatum, the largest nucleus of the basal ganglia, serves as an input port that maps cortical information. Microanatomical studies show that the striatum is a mosaic of specialized input-output structures called striosomes and regions of the surrounding matrix called the matrisomes. We have developed a computational model of the striatum using layered self-organizing maps to capture the center-surround structure seen experimentally and explain its functional significance. We believe that these structural components could build representations of state and action spaces in different environments. The striatum model is then integrated with other components of basal ganglia, making it capable of solving reinforcement learning tasks. We have proposed a biologically plausible mechanism of action-based learning where the striosome biases the matrisome activity toward a preferred action. Several studies indicate that the striatum is critical in solving context dependent problems. We build on this hypothesis and the proposed model exploits the modularity of the striatum to efficiently solve such tasks. PMID:28680395

Reinforcement Learning in Distributed Domains: Beyond Team Games

NASA Technical Reports Server (NTRS)

Wolpert, David H.; Sill, Joseph; Turner, Kagan

2000-01-01

Distributed search algorithms are crucial in dealing with large optimization problems, particularly when a centralized approach is not only impractical but infeasible. Many machine learning concepts have been applied to search algorithms in order to improve their effectiveness. In this article we present an algorithm that blends Reinforcement Learning (RL) and hill climbing directly, by using the RL signal to guide the exploration step of a hill climbing algorithm. We apply this algorithm to the domain of a constellations of communication satellites where the goal is to minimize the loss of importance weighted data. We introduce the concept of 'ghost' traffic, where correctly setting this traffic induces the satellites to act to optimize the world utility. Our results indicated that the bi-utility search introduced in this paper outperforms both traditional hill climbing algorithms and distributed RL approaches such as team games.

Lung Nodule Detection via Deep Reinforcement Learning.

PubMed

Ali, Issa; Hart, Gregory R; Gunabushanam, Gowthaman; Liang, Ying; Muhammad, Wazir; Nartowt, Bradley; Kane, Michael; Ma, Xiaomei; Deng, Jun

2018-01-01

Lung cancer is the most common cause of cancer-related death globally. As a preventive measure, the United States Preventive Services Task Force (USPSTF) recommends annual screening of high risk individuals with low-dose computed tomography (CT). The resulting volume of CT scans from millions of people will pose a significant challenge for radiologists to interpret. To fill this gap, computer-aided detection (CAD) algorithms may prove to be the most promising solution. A crucial first step in the analysis of lung cancer screening results using CAD is the detection of pulmonary nodules, which may represent early-stage lung cancer. The objective of this work is to develop and validate a reinforcement learning model based on deep artificial neural networks for early detection of lung nodules in thoracic CT images. Inspired by the AlphaGo system, our deep learning algorithm takes a raw CT image as input and views it as a collection of states, and output a classification of whether a nodule is present or not. The dataset used to train our model is the LIDC/IDRI database hosted by the lung nodule analysis (LUNA) challenge. In total, there are 888 CT scans with annotations based on agreement from at least three out of four radiologists. As a result, there are 590 individuals having one or more nodules, and 298 having none. Our training results yielded an overall accuracy of 99.1% [sensitivity 99.2%, specificity 99.1%, positive predictive value (PPV) 99.1%, negative predictive value (NPV) 99.2%]. In our test, the results yielded an overall accuracy of 64.4% (sensitivity 58.9%, specificity 55.3%, PPV 54.2%, and NPV 60.0%). These early results show promise in solving the major issue of false positives in CT screening of lung nodules, and may help to save unnecessary follow-up tests and expenditures.

Lung Nodule Detection via Deep Reinforcement Learning

PubMed Central

Ali, Issa; Hart, Gregory R.; Gunabushanam, Gowthaman; Liang, Ying; Muhammad, Wazir; Nartowt, Bradley; Kane, Michael; Ma, Xiaomei; Deng, Jun

2018-01-01

Lung cancer is the most common cause of cancer-related death globally. As a preventive measure, the United States Preventive Services Task Force (USPSTF) recommends annual screening of high risk individuals with low-dose computed tomography (CT). The resulting volume of CT scans from millions of people will pose a significant challenge for radiologists to interpret. To fill this gap, computer-aided detection (CAD) algorithms may prove to be the most promising solution. A crucial first step in the analysis of lung cancer screening results using CAD is the detection of pulmonary nodules, which may represent early-stage lung cancer. The objective of this work is to develop and validate a reinforcement learning model based on deep artificial neural networks for early detection of lung nodules in thoracic CT images. Inspired by the AlphaGo system, our deep learning algorithm takes a raw CT image as input and views it as a collection of states, and output a classification of whether a nodule is present or not. The dataset used to train our model is the LIDC/IDRI database hosted by the lung nodule analysis (LUNA) challenge. In total, there are 888 CT scans with annotations based on agreement from at least three out of four radiologists. As a result, there are 590 individuals having one or more nodules, and 298 having none. Our training results yielded an overall accuracy of 99.1% [sensitivity 99.2%, specificity 99.1%, positive predictive value (PPV) 99.1%, negative predictive value (NPV) 99.2%]. In our test, the results yielded an overall accuracy of 64.4% (sensitivity 58.9%, specificity 55.3%, PPV 54.2%, and NPV 60.0%). These early results show promise in solving the major issue of false positives in CT screening of lung nodules, and may help to save unnecessary follow-up tests and expenditures. PMID:29713615

Distributed reinforcement learning for adaptive and robust network intrusion response

NASA Astrophysics Data System (ADS)

Malialis, Kleanthis; Devlin, Sam; Kudenko, Daniel

2015-07-01

Distributed denial of service (DDoS) attacks constitute a rapidly evolving threat in the current Internet. Multiagent Router Throttling is a novel approach to defend against DDoS attacks where multiple reinforcement learning agents are installed on a set of routers and learn to rate-limit or throttle traffic towards a victim server. The focus of this paper is on online learning and scalability. We propose an approach that incorporates task decomposition, team rewards and a form of reward shaping called difference rewards. One of the novel characteristics of the proposed system is that it provides a decentralised coordinated response to the DDoS problem, thus being resilient to DDoS attacks themselves. The proposed system learns remarkably fast, thus being suitable for online learning. Furthermore, its scalability is successfully demonstrated in experiments involving 1000 learning agents. We compare our approach against a baseline and a popular state-of-the-art throttling technique from the network security literature and show that the proposed approach is more effective, adaptive to sophisticated attack rate dynamics and robust to agent failures.

Benchmarking for Bayesian Reinforcement Learning.

PubMed

Castronovo, Michael; Ernst, Damien; Couëtoux, Adrien; Fonteneau, Raphael

2016-01-01

In the Bayesian Reinforcement Learning (BRL) setting, agents try to maximise the collected rewards while interacting with their environment while using some prior knowledge that is accessed beforehand. Many BRL algorithms have already been proposed, but the benchmarks used to compare them are only relevant for specific cases. The paper addresses this problem, and provides a new BRL comparison methodology along with the corresponding open source library. In this methodology, a comparison criterion that measures the performance of algorithms on large sets of Markov Decision Processes (MDPs) drawn from some probability distributions is defined. In order to enable the comparison of non-anytime algorithms, our methodology also includes a detailed analysis of the computation time requirement of each algorithm. Our library is released with all source code and documentation: it includes three test problems, each of which has two different prior distributions, and seven state-of-the-art RL algorithms. Finally, our library is illustrated by comparing all the available algorithms and the results are discussed.

Learning to Predict Consequences as a Method of Knowledge Transfer in Reinforcement Learning.

PubMed

Chalmers, Eric; Contreras, Edgar Bermudez; Robertson, Brandon; Luczak, Artur; Gruber, Aaron

2017-04-17

The reinforcement learning (RL) paradigm allows agents to solve tasks through trial-and-error learning. To be capable of efficient, long-term learning, RL agents should be able to apply knowledge gained in the past to new tasks they may encounter in the future. The ability to predict actions' consequences may facilitate such knowledge transfer. We consider here domains where an RL agent has access to two kinds of information: agent-centric information with constant semantics across tasks, and environment-centric information, which is necessary to solve the task, but with semantics that differ between tasks. For example, in robot navigation, environment-centric information may include the robot's geographic location, while agent-centric information may include sensor readings of various nearby obstacles. We propose that these situations provide an opportunity for a very natural style of knowledge transfer, in which the agent learns to predict actions' environmental consequences using agent-centric information. These predictions contain important information about the affordances and dangers present in a novel environment, and can effectively transfer knowledge from agent-centric to environment-centric learning systems. Using several example problems including spatial navigation and network routing, we show that our knowledge transfer approach can allow faster and lower cost learning than existing alternatives.

Reinforcement learning and episodic memory in humans and animals: an integrative framework

PubMed Central

Gershman, Samuel J.; Daw, Nathaniel D.

2018-01-01

We review the psychology and neuroscience of reinforcement learning (RL), which has witnessed significant progress in the last two decades, enabled by the comprehensive experimental study of simple learning and decision-making tasks. However, the simplicity of these tasks misses important aspects of reinforcement learning in the real world: (i) State spaces are high-dimensional, continuous, and partially observable; this implies that (ii) data are relatively sparse: indeed precisely the same situation may never be encountered twice; and also that (iii) rewards depend on long-term consequences of actions in ways that violate the classical assumptions that make RL tractable. A seemingly distinct challenge is that, cognitively, these theories have largely connected with procedural and semantic memory: how knowledge about action values or world models extracted gradually from many experiences can drive choice. This misses many aspects of memory related to traces of individual events, such as episodic memory. We suggest that these two gaps are related. In particular, the computational challenges can be dealt with, in part, by endowing RL systems with episodic memory, allowing them to (i) efficiently approximate value functions over complex state spaces, (ii) learn with very little data, and (iii) bridge long-term dependencies between actions and rewards. We review the computational theory underlying this proposal and the empirical evidence to support it. Our proposal suggests that the ubiquitous and diverse roles of memory in RL may function as part of an integrated learning system. PMID:27618944

A Reinforcement Learning Model Equipped with Sensors for Generating Perception Patterns: Implementation of a Simulated Air Navigation System Using ADS-B (Automatic Dependent Surveillance-Broadcast) Technology.

PubMed

Álvarez de Toledo, Santiago; Anguera, Aurea; Barreiro, José M; Lara, Juan A; Lizcano, David

2017-01-19

Over the last few decades, a number of reinforcement learning techniques have emerged, and different reinforcement learning-based applications have proliferated. However, such techniques tend to specialize in a particular field. This is an obstacle to their generalization and extrapolation to other areas. Besides, neither the reward-punishment (r-p) learning process nor the convergence of results is fast and efficient enough. To address these obstacles, this research proposes a general reinforcement learning model. This model is independent of input and output types and based on general bioinspired principles that help to speed up the learning process. The model is composed of a perception module based on sensors whose specific perceptions are mapped as perception patterns. In this manner, similar perceptions (even if perceived at different positions in the environment) are accounted for by the same perception pattern. Additionally, the model includes a procedure that statistically associates perception-action pattern pairs depending on the positive or negative results output by executing the respective action in response to a particular perception during the learning process. To do this, the model is fitted with a mechanism that reacts positively or negatively to particular sensory stimuli in order to rate results. The model is supplemented by an action module that can be configured depending on the maneuverability of each specific agent. The model has been applied in the air navigation domain, a field with strong safety restrictions, which led us to implement a simulated system equipped with the proposed model. Accordingly, the perception sensors were based on Automatic Dependent Surveillance-Broadcast (ADS-B) technology, which is described in this paper. The results were quite satisfactory, and it outperformed traditional methods existing in the literature with respect to learning reliability and efficiency.

A Reinforcement Learning Model Equipped with Sensors for Generating Perception Patterns: Implementation of a Simulated Air Navigation System Using ADS-B (Automatic Dependent Surveillance-Broadcast) Technology

PubMed Central

Álvarez de Toledo, Santiago; Anguera, Aurea; Barreiro, José M.; Lara, Juan A.; Lizcano, David

2017-01-01

Over the last few decades, a number of reinforcement learning techniques have emerged, and different reinforcement learning-based applications have proliferated. However, such techniques tend to specialize in a particular field. This is an obstacle to their generalization and extrapolation to other areas. Besides, neither the reward-punishment (r-p) learning process nor the convergence of results is fast and efficient enough. To address these obstacles, this research proposes a general reinforcement learning model. This model is independent of input and output types and based on general bioinspired principles that help to speed up the learning process. The model is composed of a perception module based on sensors whose specific perceptions are mapped as perception patterns. In this manner, similar perceptions (even if perceived at different positions in the environment) are accounted for by the same perception pattern. Additionally, the model includes a procedure that statistically associates perception-action pattern pairs depending on the positive or negative results output by executing the respective action in response to a particular perception during the learning process. To do this, the model is fitted with a mechanism that reacts positively or negatively to particular sensory stimuli in order to rate results. The model is supplemented by an action module that can be configured depending on the maneuverability of each specific agent. The model has been applied in the air navigation domain, a field with strong safety restrictions, which led us to implement a simulated system equipped with the proposed model. Accordingly, the perception sensors were based on Automatic Dependent Surveillance-Broadcast (ADS-B) technology, which is described in this paper. The results were quite satisfactory, and it outperformed traditional methods existing in the literature with respect to learning reliability and efficiency. PMID:28106849

Dose Dependent Dopaminergic Modulation of Reward-Based Learning in Parkinson's Disease

ERIC Educational Resources Information Center

van Wouwe, N. C.; Ridderinkhof, K. R.; Band, G. P. H.; van den Wildenberg, W. P. M.; Wylie, S. A.

2012-01-01

Learning to select optimal behavior in new and uncertain situations is a crucial aspect of living and requires the ability to quickly associate stimuli with actions that lead to rewarding outcomes. Mathematical models of reinforcement-based learning to select rewarding actions distinguish between (1) the formation of stimulus-action-reward…

Neural mechanisms of reinforcement learning in unmedicated patients with major depressive disorder.

PubMed

Rothkirch, Marcus; Tonn, Jonas; Köhler, Stephan; Sterzer, Philipp

2017-04-01

According to current concepts, major depressive disorder is strongly related to dysfunctional neural processing of motivational information, entailing impairments in reinforcement learning. While computational modelling can reveal the precise nature of neural learning signals, it has not been used to study learning-related neural dysfunctions in unmedicated patients with major depressive disorder so far. We thus aimed at comparing the neural coding of reward and punishment prediction errors, representing indicators of neural learning-related processes, between unmedicated patients with major depressive disorder and healthy participants. To this end, a group of unmedicated patients with major depressive disorder (n = 28) and a group of age- and sex-matched healthy control participants (n = 30) completed an instrumental learning task involving monetary gains and losses during functional magnetic resonance imaging. The two groups did not differ in their learning performance. Patients and control participants showed the same level of prediction error-related activity in the ventral striatum and the anterior insula. In contrast, neural coding of reward prediction errors in the medial orbitofrontal cortex was reduced in patients. Moreover, neural reward prediction error signals in the medial orbitofrontal cortex and ventral striatum showed negative correlations with anhedonia severity. Using a standard instrumental learning paradigm we found no evidence for an overall impairment of reinforcement learning in medication-free patients with major depressive disorder. Importantly, however, the attenuated neural coding of reward in the medial orbitofrontal cortex and the relation between anhedonia and reduced reward prediction error-signalling in the medial orbitofrontal cortex and ventral striatum likely reflect an impairment in experiencing pleasure from rewarding events as a key mechanism of anhedonia in major depressive disorder. © The Author (2017). Published by Oxford

Reinforcement Learning Models and Their Neural Correlates: An Activation Likelihood Estimation Meta-Analysis

PubMed Central

Kumar, Poornima; Eickhoff, Simon B.; Dombrovski, Alexandre Y.

2015-01-01

Reinforcement learning describes motivated behavior in terms of two abstract signals. The representation of discrepancies between expected and actual rewards/punishments – prediction error – is thought to update the expected value of actions and predictive stimuli. Electrophysiological and lesion studies suggest that mesostriatal prediction error signals control behavior through synaptic modification of cortico-striato-thalamic networks. Signals in the ventromedial prefrontal and orbitofrontal cortex are implicated in representing expected value. To obtain unbiased maps of these representations in the human brain, we performed a meta-analysis of functional magnetic resonance imaging studies that employed algorithmic reinforcement learning models, across a variety of experimental paradigms. We found that the ventral striatum (medial and lateral) and midbrain/thalamus represented reward prediction errors, consistent with animal studies. Prediction error signals were also seen in the frontal operculum/insula, particularly for social rewards. In Pavlovian studies, striatal prediction error signals extended into the amygdala, while instrumental tasks engaged the caudate. Prediction error maps were sensitive to the model-fitting procedure (fixed or individually-estimated) and to the extent of spatial smoothing. A correlate of expected value was found in a posterior region of the ventromedial prefrontal cortex, caudal and medial to the orbitofrontal regions identified in animal studies. These findings highlight a reproducible motif of reinforcement learning in the cortico-striatal loops and identify methodological dimensions that may influence the reproducibility of activation patterns across studies. PMID:25665667

Safe Exploration Algorithms for Reinforcement Learning Controllers.

PubMed

Mannucci, Tommaso; van Kampen, Erik-Jan; de Visser, Cornelis; Chu, Qiping

2018-04-01

Self-learning approaches, such as reinforcement learning, offer new possibilities for autonomous control of uncertain or time-varying systems. However, exploring an unknown environment under limited prediction capabilities is a challenge for a learning agent. If the environment is dangerous, free exploration can result in physical damage or in an otherwise unacceptable behavior. With respect to existing methods, the main contribution of this paper is the definition of a new approach that does not require global safety functions, nor specific formulations of the dynamics or of the environment, but relies on interval estimation of the dynamics of the agent during the exploration phase, assuming a limited capability of the agent to perceive the presence of incoming fatal states. Two algorithms are presented with this approach. The first is the Safety Handling Exploration with Risk Perception Algorithm (SHERPA), which provides safety by individuating temporary safety functions, called backups. SHERPA is shown in a simulated, simplified quadrotor task, for which dangerous states are avoided. The second algorithm, denominated OptiSHERPA, can safely handle more dynamically complex systems for which SHERPA is not sufficient through the use of safety metrics. An application of OptiSHERPA is simulated on an aircraft altitude control task.

Reinforcement learning and counterfactual reasoning explain adaptive behavior in a changing environment.

PubMed

Zhang, Yunfeng; Paik, Jaehyon; Pirolli, Peter

2015-04-01

Animals routinely adapt to changes in the environment in order to survive. Though reinforcement learning may play a role in such adaptation, it is not clear that it is the only mechanism involved, as it is not well suited to producing rapid, relatively immediate changes in strategies in response to environmental changes. This research proposes that counterfactual reasoning might be an additional mechanism that facilitates change detection. An experiment is conducted in which a task state changes over time and the participants had to detect the changes in order to perform well and gain monetary rewards. A cognitive model is constructed that incorporates reinforcement learning with counterfactual reasoning to help quickly adjust the utility of task strategies in response to changes. The results show that the model can accurately explain human data and that counterfactual reasoning is key to reproducing the various effects observed in this change detection paradigm. Copyright © 2015 Cognitive Science Society, Inc.

Reinforcement learning techniques for controlling resources in power networks

NASA Astrophysics Data System (ADS)

Kowli, Anupama Sunil

As power grids transition towards increased reliance on renewable generation, energy storage and demand response resources, an effective control architecture is required to harness the full functionalities of these resources. There is a critical need for control techniques that recognize the unique characteristics of the different resources and exploit the flexibility afforded by them to provide ancillary services to the grid. The work presented in this dissertation addresses these needs. Specifically, new algorithms are proposed, which allow control synthesis in settings wherein the precise distribution of the uncertainty and its temporal statistics are not known. These algorithms are based on recent developments in Markov decision theory, approximate dynamic programming and reinforcement learning. They impose minimal assumptions on the system model and allow the control to be "learned" based on the actual dynamics of the system. Furthermore, they can accommodate complex constraints such as capacity and ramping limits on generation resources, state-of-charge constraints on storage resources, comfort-related limitations on demand response resources and power flow limits on transmission lines. Numerical studies demonstrating applications of these algorithms to practical control problems in power systems are discussed. Results demonstrate how the proposed control algorithms can be used to improve the performance and reduce the computational complexity of the economic dispatch mechanism in a power network. We argue that the proposed algorithms are eminently suitable to develop operational decision-making tools for large power grids with many resources and many sources of uncertainty.

Dopaminergic control of motivation and reinforcement learning: a closed-circuit account for reward-oriented behavior.

PubMed

Morita, Kenji; Morishima, Mieko; Sakai, Katsuyuki; Kawaguchi, Yasuo

2013-05-15

Humans and animals take actions quickly when they expect that the actions lead to reward, reflecting their motivation. Injection of dopamine receptor antagonists into the striatum has been shown to slow such reward-seeking behavior, suggesting that dopamine is involved in the control of motivational processes. Meanwhile, neurophysiological studies have revealed that phasic response of dopamine neurons appears to represent reward prediction error, indicating that dopamine plays central roles in reinforcement learning. However, previous attempts to elucidate the mechanisms of these dopaminergic controls have not fully explained how the motivational and learning aspects are related and whether they can be understood by the way the activity of dopamine neurons itself is controlled by their upstream circuitries. To address this issue, we constructed a closed-circuit model of the corticobasal ganglia system based on recent findings regarding intracortical and corticostriatal circuit architectures. Simulations show that the model could reproduce the observed distinct motivational effects of D1- and D2-type dopamine receptor antagonists. Simultaneously, our model successfully explains the dopaminergic representation of reward prediction error as observed in behaving animals during learning tasks and could also explain distinct choice biases induced by optogenetic stimulation of the D1 and D2 receptor-expressing striatal neurons. These results indicate that the suggested roles of dopamine in motivational control and reinforcement learning can be understood in a unified manner through a notion that the indirect pathway of the basal ganglia represents the value of states/actions at a previous time point, an empirically driven key assumption of our model.

Switching Reinforcement Learning for Continuous Action Space

NASA Astrophysics Data System (ADS)

Nagayoshi, Masato; Murao, Hajime; Tamaki, Hisashi

Reinforcement Learning (RL) attracts much attention as a technique of realizing computational intelligence such as adaptive and autonomous decentralized systems. In general, however, it is not easy to put RL into practical use. This difficulty includes a problem of designing a suitable action space of an agent, i.e., satisfying two requirements in trade-off: (i) to keep the characteristics (or structure) of an original search space as much as possible in order to seek strategies that lie close to the optimal, and (ii) to reduce the search space as much as possible in order to expedite the learning process. In order to design a suitable action space adaptively, we propose switching RL model to mimic a process of an infant's motor development in which gross motor skills develop before fine motor skills. Then, a method for switching controllers is constructed by introducing and referring to the “entropy”. Further, through computational experiments by using robot navigation problems with one and two-dimensional continuous action space, the validity of the proposed method has been confirmed.

Self-regulation of the anterior insula: Reinforcement learning using real-time fMRI neurofeedback.

PubMed

Lawrence, Emma J; Su, Li; Barker, Gareth J; Medford, Nick; Dalton, Jeffrey; Williams, Steve C R; Birbaumer, Niels; Veit, Ralf; Ranganatha, Sitaram; Bodurka, Jerzy; Brammer, Michael; Giampietro, Vincent; David, Anthony S

2014-03-01

The anterior insula (AI) plays a key role in affective processing, and insular dysfunction has been noted in several clinical conditions. Real-time functional MRI neurofeedback (rtfMRI-NF) provides a means of helping people learn to self-regulate activation in this brain region. Using the Blood Oxygenated Level Dependant (BOLD) signal from the right AI (RAI) as neurofeedback, we trained participants to increase RAI activation. In contrast, another group of participants was shown 'control' feedback from another brain area. Pre- and post-training affective probes were shown, with subjective ratings and skin conductance response (SCR) measured. We also investigated a reward-related reinforcement learning model of rtfMRI-NF. In contrast to the controls, we hypothesised a positive linear increase in RAI activation in participants shown feedback from this region, alongside increases in valence ratings and SCR to affective probes. Hypothesis-driven analyses showed a significant interaction between the RAI/control neurofeedback groups and the effect of self-regulation. Whole-brain analyses revealed a significant linear increase in RAI activation across four training runs in the group who received feedback from RAI. Increased activation was also observed in the caudate body and thalamus, likely representing feedback-related learning. No positive linear trend was observed in the RAI in the group receiving control feedback, suggesting that these data are not a general effect of cognitive strategy or control feedback. The control group did, however, show diffuse activation across the putamen, caudate and posterior insula which may indicate the representation of false feedback. No significant training-related behavioural differences were observed for valence ratings, or SCR. In addition, correlational analyses based on a reinforcement learning model showed that the dorsal anterior cingulate cortex underpinned learning in both groups. In summary, these data demonstrate that it

Corticostriatal circuit mechanisms of value-based action selection: Implementation of reinforcement learning algorithms and beyond.

PubMed

Morita, Kenji; Jitsev, Jenia; Morrison, Abigail

2016-09-15

Value-based action selection has been suggested to be realized in the corticostriatal local circuits through competition among neural populations. In this article, we review theoretical and experimental studies that have constructed and verified this notion, and provide new perspectives on how the local-circuit selection mechanisms implement reinforcement learning (RL) algorithms and computations beyond them. The striatal neurons are mostly inhibitory, and lateral inhibition among them has been classically proposed to realize "Winner-Take-All (WTA)" selection of the maximum-valued action (i.e., 'max' operation). Although this view has been challenged by the revealed weakness, sparseness, and asymmetry of lateral inhibition, which suggest more complex dynamics, WTA-like competition could still occur on short time scales. Unlike the striatal circuit, the cortical circuit contains recurrent excitation, which may enable retention or temporal integration of information and probabilistic "soft-max" selection. The striatal "max" circuit and the cortical "soft-max" circuit might co-implement an RL algorithm called Q-learning; the cortical circuit might also similarly serve for other algorithms such as SARSA. In these implementations, the cortical circuit presumably sustains activity representing the executed action, which negatively impacts dopamine neurons so that they can calculate reward-prediction-error. Regarding the suggested more complex dynamics of striatal, as well as cortical, circuits on long time scales, which could be viewed as a sequence of short WTA fragments, computational roles remain open: such a sequence might represent (1) sequential state-action-state transitions, constituting replay or simulation of the internal model, (2) a single state/action by the whole trajectory, or (3) probabilistic sampling of state/action. Copyright © 2016. Published by Elsevier B.V.

Confirmation bias in human reinforcement learning: Evidence from counterfactual feedback processing

PubMed Central

Lefebvre, Germain; Blakemore, Sarah-Jayne

2017-01-01

Previous studies suggest that factual learning, that is, learning from obtained outcomes, is biased, such that participants preferentially take into account positive, as compared to negative, prediction errors. However, whether or not the prediction error valence also affects counterfactual learning, that is, learning from forgone outcomes, is unknown. To address this question, we analysed the performance of two groups of participants on reinforcement learning tasks using a computational model that was adapted to test if prediction error valence influences learning. We carried out two experiments: in the factual learning experiment, participants learned from partial feedback (i.e., the outcome of the chosen option only); in the counterfactual learning experiment, participants learned from complete feedback information (i.e., the outcomes of both the chosen and unchosen option were displayed). In the factual learning experiment, we replicated previous findings of a valence-induced bias, whereby participants learned preferentially from positive, relative to negative, prediction errors. In contrast, for counterfactual learning, we found the opposite valence-induced bias: negative prediction errors were preferentially taken into account, relative to positive ones. When considering valence-induced bias in the context of both factual and counterfactual learning, it appears that people tend to preferentially take into account information that confirms their current choice. PMID:28800597

Confirmation bias in human reinforcement learning: Evidence from counterfactual feedback processing.

PubMed

Palminteri, Stefano; Lefebvre, Germain; Kilford, Emma J; Blakemore, Sarah-Jayne

2017-08-01

Previous studies suggest that factual learning, that is, learning from obtained outcomes, is biased, such that participants preferentially take into account positive, as compared to negative, prediction errors. However, whether or not the prediction error valence also affects counterfactual learning, that is, learning from forgone outcomes, is unknown. To address this question, we analysed the performance of two groups of participants on reinforcement learning tasks using a computational model that was adapted to test if prediction error valence influences learning. We carried out two experiments: in the factual learning experiment, participants learned from partial feedback (i.e., the outcome of the chosen option only); in the counterfactual learning experiment, participants learned from complete feedback information (i.e., the outcomes of both the chosen and unchosen option were displayed). In the factual learning experiment, we replicated previous findings of a valence-induced bias, whereby participants learned preferentially from positive, relative to negative, prediction errors. In contrast, for counterfactual learning, we found the opposite valence-induced bias: negative prediction errors were preferentially taken into account, relative to positive ones. When considering valence-induced bias in the context of both factual and counterfactual learning, it appears that people tend to preferentially take into account information that confirms their current choice.

Within- and across-trial dynamics of human EEG reveal cooperative interplay between reinforcement learning and working memory.

PubMed

Collins, Anne G E; Frank, Michael J

2018-03-06

Learning from rewards and punishments is essential to survival and facilitates flexible human behavior. It is widely appreciated that multiple cognitive and reinforcement learning systems contribute to decision-making, but the nature of their interactions is elusive. Here, we leverage methods for extracting trial-by-trial indices of reinforcement learning (RL) and working memory (WM) in human electro-encephalography to reveal single-trial computations beyond that afforded by behavior alone. Neural dynamics confirmed that increases in neural expectation were predictive of reduced neural surprise in the following feedback period, supporting central tenets of RL models. Within- and cross-trial dynamics revealed a cooperative interplay between systems for learning, in which WM contributes expectations to guide RL, despite competition between systems during choice. Together, these results provide a deeper understanding of how multiple neural systems interact for learning and decision-making and facilitate analysis of their disruption in clinical populations.

Stochastic Reinforcement Benefits Skill Acquisition

ERIC Educational Resources Information Center

Dayan, Eran; Averbeck, Bruno B.; Richmond, Barry J.; Cohen, Leonardo G.

2014-01-01

Learning complex skills is driven by reinforcement, which facilitates both online within-session gains and retention of the acquired skills. Yet, in ecologically relevant situations, skills are often acquired when mapping between actions and rewarding outcomes is unknown to the learning agent, resulting in reinforcement schedules of a stochastic…

Does overall reinforcer rate affect discrimination of time-based contingencies?

PubMed

Cowie, Sarah; Davison, Michael; Blumhardt, Luca; Elliffe, Douglas

2016-05-01

Overall reinforcer rate appears to affect choice. The mechanism for such an effect is uncertain, but may relate to reinforcer rate changing the discrimination of the relation between stimuli and reinforcers. We assessed whether a quantitative model based on a stimulus-control approach could be used to account for the effects of overall reinforcer rate on choice under changing time-based contingencies. On a two-key concurrent schedule, the likely availability of a reinforcer reversed when a fixed time had elapsed since the last reinforcer, and the overall reinforcer rate was varied across conditions. Changes in the overall reinforcer rate produced a change in response bias, and some indication of a change in discrimination. These changes in bias and discrimination always occurred quickly, usually within the first session of a condition. The stimulus-control approach provided an excellent account of the data, suggesting that changes in overall reinforcer rate affect choice because they alter the frequency of reinforcers obtained at different times, or in different stimulus contexts, and thus change the discriminated relation between stimuli and reinforcers. These findings support the notion that temporal and spatial discriminations can be understood in terms of discrimination of reinforcers across time and space. © 2016 Society for the Experimental Analysis of Behavior.

Mechanisms of Hierarchical Reinforcement Learning in Corticostriatal Circuits 1: Computational Analysis

PubMed Central

Badre, David

2012-01-01

Growing evidence suggests that the prefrontal cortex (PFC) is organized hierarchically, with more anterior regions having increasingly abstract representations. How does this organization support hierarchical cognitive control and the rapid discovery of abstract action rules? We present computational models at different levels of description. A neural circuit model simulates interacting corticostriatal circuits organized hierarchically. In each circuit, the basal ganglia gate frontal actions, with some striatal units gating the inputs to PFC and others gating the outputs to influence response selection. Learning at all of these levels is accomplished via dopaminergic reward prediction error signals in each corticostriatal circuit. This functionality allows the system to exhibit conditional if–then hypothesis testing and to learn rapidly in environments with hierarchical structure. We also develop a hybrid Bayesian-reinforcement learning mixture of experts (MoE) model, which can estimate the most likely hypothesis state of individual participants based on their observed sequence of choices and rewards. This model yields accurate probabilistic estimates about which hypotheses are attended by manipulating attentional states in the generative neural model and recovering them with the MoE model. This 2-pronged modeling approach leads to multiple quantitative predictions that are tested with functional magnetic resonance imaging in the companion paper. PMID:21693490

Developing PFC representations using reinforcement learning.

PubMed

Reynolds, Jeremy R; O'Reilly, Randall C

2009-12-01

From both functional and biological considerations, it is widely believed that action production, planning, and goal-oriented behaviors supported by the frontal cortex are organized hierarchically [Fuster (1991); Koechlin, E., Ody, C., & Kouneiher, F. (2003). Neuroscience: The architecture of cognitive control in the human prefrontal cortex. Science, 424, 1181-1184; Miller, G. A., Galanter, E., & Pribram, K. H. (1960). Plans and the structure of behavior. New York: Holt]. However, the nature of the different levels of the hierarchy remains unclear, and little attention has been paid to the origins of such a hierarchy. We address these issues through biologically-inspired computational models that develop representations through reinforcement learning. We explore several different factors in these models that might plausibly give rise to a hierarchical organization of representations within the PFC, including an initial connectivity hierarchy within PFC, a hierarchical set of connections between PFC and subcortical structures controlling it, and differential synaptic plasticity schedules. Simulation results indicate that architectural constraints contribute to the segregation of different types of representations, and that this segregation facilitates learning. These findings are consistent with the idea that there is a functional hierarchy in PFC, as captured in our earlier computational models of PFC function and a growing body of empirical data.

Market Model for Resource Allocation in Emerging Sensor Networks with Reinforcement Learning

PubMed Central

Zhang, Yue; Song, Bin; Zhang, Ying; Du, Xiaojiang; Guizani, Mohsen

2016-01-01

Emerging sensor networks (ESNs) are an inevitable trend with the development of the Internet of Things (IoT), and intend to connect almost every intelligent device. Therefore, it is critical to study resource allocation in such an environment, due to the concern of efficiency, especially when resources are limited. By viewing ESNs as multi-agent environments, we model them with an agent-based modelling (ABM) method and deal with resource allocation problems with market models, after describing users’ patterns. Reinforcement learning methods are introduced to estimate users’ patterns and verify the outcomes in our market models. Experimental results show the efficiency of our methods, which are also capable of guiding topology management. PMID:27916841

Reinforcement Learning and Episodic Memory in Humans and Animals: An Integrative Framework.

PubMed

Gershman, Samuel J; Daw, Nathaniel D

2017-01-03

We review the psychology and neuroscience of reinforcement learning (RL), which has experienced significant progress in the past two decades, enabled by the comprehensive experimental study of simple learning and decision-making tasks. However, one challenge in the study of RL is computational: The simplicity of these tasks ignores important aspects of reinforcement learning in the real world: (a) State spaces are high-dimensional, continuous, and partially observable; this implies that (b) data are relatively sparse and, indeed, precisely the same situation may never be encountered twice; furthermore, (c) rewards depend on the long-term consequences of actions in ways that violate the classical assumptions that make RL tractable. A seemingly distinct challenge is that, cognitively, theories of RL have largely involved procedural and semantic memory, the way in which knowledge about action values or world models extracted gradually from many experiences can drive choice. This focus on semantic memory leaves out many aspects of memory, such as episodic memory, related to the traces of individual events. We suggest that these two challenges are related. The computational challenge can be dealt with, in part, by endowing RL systems with episodic memory, allowing them to (a) efficiently approximate value functions over complex state spaces, (b) learn with very little data, and (c) bridge long-term dependencies between actions and rewards. We review the computational theory underlying this proposal and the empirical evidence to support it. Our proposal suggests that the ubiquitous and diverse roles of memory in RL may function as part of an integrated learning system.

The left hemisphere learns what is right: Hemispatial reward learning depends on reinforcement learning processes in the contralateral hemisphere.

PubMed

Aberg, Kristoffer Carl; Doell, Kimberly Crystal; Schwartz, Sophie

2016-08-01

Orienting biases refer to consistent, trait-like direction of attention or locomotion toward one side of space. Recent studies suggest that such hemispatial biases may determine how well people memorize information presented in the left or right hemifield. Moreover, lesion studies indicate that learning rewarded stimuli in one hemispace depends on the integrity of the contralateral striatum. However, the exact neural and computational mechanisms underlying the influence of individual orienting biases on reward learning remain unclear. Because reward-based behavioural adaptation depends on the dopaminergic system and prediction error (PE) encoding in the ventral striatum, we hypothesized that hemispheric asymmetries in dopamine (DA) function may determine individual spatial biases in reward learning. To test this prediction, we acquired fMRI in 33 healthy human participants while they performed a lateralized reward task. Learning differences between hemispaces were assessed by presenting stimuli, assigned to different reward probabilities, to the left or right of central fixation, i.e. presented in the left or right visual hemifield. Hemispheric differences in DA function were estimated through differential fMRI responses to positive vs. negative feedback in the left vs. right ventral striatum, and a computational approach was used to identify the neural correlates of PEs. Our results show that spatial biases favoring reward learning in the right (vs. left) hemifield were associated with increased reward responses in the left hemisphere and relatively better neural encoding of PEs for stimuli presented in the right (vs. left) hemifield. These findings demonstrate that trait-like spatial biases implicate hemisphere-specific learning mechanisms, with individual differences between hemispheres contributing to reinforcing spatial biases. Copyright © 2016 Elsevier Ltd. All rights reserved.

Novelty and Inductive Generalization in Human Reinforcement Learning.

PubMed

Gershman, Samuel J; Niv, Yael

2015-07-01

In reinforcement learning (RL), a decision maker searching for the most rewarding option is often faced with the question: What is the value of an option that has never been tried before? One way to frame this question is as an inductive problem: How can I generalize my previous experience with one set of options to a novel option? We show how hierarchical Bayesian inference can be used to solve this problem, and we describe an equivalence between the Bayesian model and temporal difference learning algorithms that have been proposed as models of RL in humans and animals. According to our view, the search for the best option is guided by abstract knowledge about the relationships between different options in an environment, resulting in greater search efficiency compared to traditional RL algorithms previously applied to human cognition. In two behavioral experiments, we test several predictions of our model, providing evidence that humans learn and exploit structured inductive knowledge to make predictions about novel options. In light of this model, we suggest a new interpretation of dopaminergic responses to novelty. Copyright © 2015 Cognitive Science Society, Inc.

Learning Based Bidding Strategy for HVAC Systems in Double Auction Retail Energy Markets

DOE Office of Scientific and Technical Information (OSTI.GOV)

Sun, Yannan; Somani, Abhishek; Carroll, Thomas E.

In this paper, a bidding strategy is proposed using reinforcement learning for HVAC systems in a double auction market. The bidding strategy does not require a specific model-based representation of behavior, i.e., a functional form to translate indoor house temperatures into bid prices. The results from reinforcement learning based approach are compared with the HVAC bidding approach used in the AEP gridSMART® smart grid demonstration project and it is shown that the model-free (learning based) approach tracks well the results from the model-based behavior. Successful use of model-free approaches to represent device-level economic behavior may help develop similar approaches tomore » represent behavior of more complex devices or groups of diverse devices, such as in a building. Distributed control requires an understanding of decision making processes of intelligent agents so that appropriate mechanisms may be developed to control and coordinate their responses, and model-free approaches to represent behavior will be extremely useful in that quest.« less

Dynamic route and departure time choice model based on self-adaptive reference point and reinforcement learning

NASA Astrophysics Data System (ADS)

Li, Xue-yan; Li, Xue-mei; Yang, Lingrun; Li, Jing

2018-07-01

Most of the previous studies on dynamic traffic assignment are based on traditional analytical framework, for instance, the idea of Dynamic User Equilibrium has been widely used in depicting both the route choice and the departure time choice. However, some recent studies have demonstrated that the dynamic traffic flow assignment largely depends on travelers' rationality degree, travelers' heterogeneity and what the traffic information the travelers have. In this paper, we develop a new self-adaptive multi agent model to depict travelers' behavior in Dynamic Traffic Assignment. We use Cumulative Prospect Theory with heterogeneous reference points to illustrate travelers' bounded rationality. We use reinforcement-learning model to depict travelers' route and departure time choosing behavior under the condition of imperfect information. We design the evolution rule of travelers' expected arrival time and the algorithm of traffic flow assignment. Compared with the traditional model, the self-adaptive multi agent model we proposed in this paper can effectively help travelers avoid the rush hour. Finally, we report and analyze the effect of travelers' group behavior on the transportation system, and give some insights into the relation between travelers' group behavior and the performance of transportation system.

Stochastic abstract policies: generalizing knowledge to improve reinforcement learning.

PubMed

Koga, Marcelo L; Freire, Valdinei; Costa, Anna H R

2015-01-01

Reinforcement learning (RL) enables an agent to learn behavior by acquiring experience through trial-and-error interactions with a dynamic environment. However, knowledge is usually built from scratch and learning to behave may take a long time. Here, we improve the learning performance by leveraging prior knowledge; that is, the learner shows proper behavior from the beginning of a target task, using the knowledge from a set of known, previously solved, source tasks. In this paper, we argue that building stochastic abstract policies that generalize over past experiences is an effective way to provide such improvement and this generalization outperforms the current practice of using a library of policies. We achieve that contributing with a new algorithm, AbsProb-PI-multiple and a framework for transferring knowledge represented as a stochastic abstract policy in new RL tasks. Stochastic abstract policies offer an effective way to encode knowledge because the abstraction they provide not only generalizes solutions but also facilitates extracting the similarities among tasks. We perform experiments in a robotic navigation environment and analyze the agent's behavior throughout the learning process and also assess the transfer ratio for different amounts of source tasks. We compare our method with the transfer of a library of policies, and experiments show that the use of a generalized policy produces better results by more effectively guiding the agent when learning a target task.

Benchmarking for Bayesian Reinforcement Learning

PubMed Central

Ernst, Damien; Couëtoux, Adrien

2016-01-01

In the Bayesian Reinforcement Learning (BRL) setting, agents try to maximise the collected rewards while interacting with their environment while using some prior knowledge that is accessed beforehand. Many BRL algorithms have already been proposed, but the benchmarks used to compare them are only relevant for specific cases. The paper addresses this problem, and provides a new BRL comparison methodology along with the corresponding open source library. In this methodology, a comparison criterion that measures the performance of algorithms on large sets of Markov Decision Processes (MDPs) drawn from some probability distributions is defined. In order to enable the comparison of non-anytime algorithms, our methodology also includes a detailed analysis of the computation time requirement of each algorithm. Our library is released with all source code and documentation: it includes three test problems, each of which has two different prior distributions, and seven state-of-the-art RL algorithms. Finally, our library is illustrated by comparing all the available algorithms and the results are discussed. PMID:27304891

Changes in the stimulus-preceding negativity and lateralized readiness potential during reinforcement learning.

PubMed

Ren, Xi; Valle-Inclán, Fernando; Tukaiev, Sergii; Hackley, Steven A

2017-07-01

According to reinforcement learning theory, dopamine-dependent anticipatory processes play a critical role in learning from action outcomes such as feedback or reward. To better understand outcome anticipation, we examined variation in slow cortical potentials and assessed their changes over the course of motor-skill acquisition. Healthy young adults learned a series of precisely timed, key press sequences. Feedback was delivered at a delay of either 2.5 or 8 s, to encourage use of either the striatally mediated, habit learning system or the hippocampus-dependent, episodic memory system, respectively. During the 2.5-s delay, the stimulus-preceding negativity (SPN) was shown to decline in amplitude across trials, confirming previous results from a perceptual categorization task (Morís, Luque, & Rodríguez-Fornells, 2013). This falsifies the hypothesis that SPN reflects specific outcome predictions, on the assumption that the ability to make such predictions should improve as a task is mastered. An SPN was also evident during the 8-s delay, but it increased in amplitude across trials. At the conclusion of the 8-s but not the 2.5-s prefeedback interval, a reversed-polarity lateralized readiness potential (LRP) was noted. It was suggested that this might indicate maintenance of an action representation for comparison with the feedback display. If so, this would constitute the first direct psychophysiological evidence for a popular hypothetical construct in quantitative models of reinforcement learning, the so-called eligibility trace. © 2017 Society for Psychophysiological Research.

Striatum-medial prefrontal cortex connectivity predicts developmental changes in reinforcement learning.

PubMed

van den Bos, Wouter; Cohen, Michael X; Kahnt, Thorsten; Crone, Eveline A

2012-06-01

During development, children improve in learning from feedback to adapt their behavior. However, it is still unclear which neural mechanisms might underlie these developmental changes. In the current study, we used a reinforcement learning model to investigate neurodevelopmental changes in the representation and processing of learning signals. Sixty-seven healthy volunteers between ages 8 and 22 (children: 8-11 years, adolescents: 13-16 years, and adults: 18-22 years) performed a probabilistic learning task while in a magnetic resonance imaging scanner. The behavioral data demonstrated age differences in learning parameters with a stronger impact of negative feedback on expected value in children. Imaging data revealed that the neural representation of prediction errors was similar across age groups, but functional connectivity between the ventral striatum and the medial prefrontal cortex changed as a function of age. Furthermore, the connectivity strength predicted the tendency to alter expectations after receiving negative feedback. These findings suggest that the underlying mechanisms of developmental changes in learning are not related to differences in the neural representation of learning signals per se but rather in how learning signals are used to guide behavior and expectations.

Reinforcement learning produces dominant strategies for the Iterated Prisoner's Dilemma.

PubMed

Harper, Marc; Knight, Vincent; Jones, Martin; Koutsovoulos, Georgios; Glynatsi, Nikoleta E; Campbell, Owen

2017-01-01

We present tournament results and several powerful strategies for the Iterated Prisoner's Dilemma created using reinforcement learning techniques (evolutionary and particle swarm algorithms). These strategies are trained to perform well against a corpus of over 170 distinct opponents, including many well-known and classic strategies. All the trained strategies win standard tournaments against the total collection of other opponents. The trained strategies and one particular human made designed strategy are the top performers in noisy tournaments also.

The nature of sexual reinforcement.

PubMed Central

Crawford, L L; Holloway, K S; Domjan, M

1993-01-01

Sexual reinforcers are not part of a regulatory system involved in the maintenance of critical metabolic processes, they differ for males and females, they differ as a function of species and mating system, and they show ontogenetic and seasonal changes related to endocrine conditions. Exposure to a member of the opposite sex without copulation can be sufficient for sexual reinforcement. However, copulatory access is a stronger reinforcer, and copulatory opportunity can serve to enhance the reinforcing efficacy of stimulus features of a sexual partner. Conversely, under certain conditions, noncopulatory exposure serves to decrease reinforcer efficacy. Many common learning phenomena such as acquisition, extinction, discrimination learning, second-order conditioning, and latent inhibition have been demonstrated in sexual conditioning. These observations extend the generality of findings obtained with more conventional reinforcers, but the mechanisms of these effects and their gender and species specificity remain to be explored. PMID:8354970

Frontal Theta Links Prediction Errors to Behavioral Adaptation in Reinforcement Learning

PubMed Central

Cavanagh, James F.; Frank, Michael J.; Klein, Theresa J.; Allen, John J.B.

2009-01-01

Investigations into action monitoring have consistently detailed a fronto-central voltage deflection in the Event-Related Potential (ERP) following the presentation of negatively valenced feedback, sometimes termed the Feedback Related Negativity (FRN). The FRN has been proposed to reflect a neural response to prediction errors during reinforcement learning, yet the single trial relationship between neural activity and the quanta of expectation violation remains untested. Although ERP methods are not well suited to single trial analyses, the FRN has been associated with theta band oscillatory perturbations in the medial prefrontal cortex. Medio-frontal theta oscillations have been previously associated with expectation violation and behavioral adaptation and are well suited to single trial analysis. Here, we recorded EEG activity during a probabilistic reinforcement learning task and fit the performance data to an abstract computational model (Q-learning) for calculation of single-trial reward prediction errors. Single-trial theta oscillatory activities following feedback were investigated within the context of expectation (prediction error) and adaptation (subsequent reaction time change). Results indicate that interactive medial and lateral frontal theta activities reflect the degree of negative and positive reward prediction error in the service of behavioral adaptation. These different brain areas use prediction error calculations for different behavioral adaptations: with medial frontal theta reflecting the utilization of prediction errors for reaction time slowing (specifically following errors), but lateral frontal theta reflecting prediction errors leading to working memory-related reaction time speeding for the correct choice. PMID:19969093

Cognitive components underpinning the development of model-based learning.

PubMed

Potter, Tracey C S; Bryce, Nessa V; Hartley, Catherine A

2017-06-01

Reinforcement learning theory distinguishes "model-free" learning, which fosters reflexive repetition of previously rewarded actions, from "model-based" learning, which recruits a mental model of the environment to flexibly select goal-directed actions. Whereas model-free learning is evident across development, recruitment of model-based learning appears to increase with age. However, the cognitive processes underlying the development of model-based learning remain poorly characterized. Here, we examined whether age-related differences in cognitive processes underlying the construction and flexible recruitment of mental models predict developmental increases in model-based choice. In a cohort of participants aged 9-25, we examined whether the abilities to infer sequential regularities in the environment ("statistical learning"), maintain information in an active state ("working memory") and integrate distant concepts to solve problems ("fluid reasoning") predicted age-related improvements in model-based choice. We found that age-related improvements in statistical learning performance did not mediate the relationship between age and model-based choice. Ceiling performance on our working memory assay prevented examination of its contribution to model-based learning. However, age-related improvements in fluid reasoning statistically mediated the developmental increase in the recruitment of a model-based strategy. These findings suggest that gradual development of fluid reasoning may be a critical component process underlying the emergence of model-based learning. Copyright © 2016 The Authors. Published by Elsevier Ltd.. All rights reserved.

Integral reinforcement learning for continuous-time input-affine nonlinear systems with simultaneous invariant explorations.

PubMed

Lee, Jae Young; Park, Jin Bae; Choi, Yoon Ho

2015-05-01

This paper focuses on a class of reinforcement learning (RL) algorithms, named integral RL (I-RL), that solve continuous-time (CT) nonlinear optimal control problems with input-affine system dynamics. First, we extend the concepts of exploration, integral temporal difference, and invariant admissibility to the target CT nonlinear system that is governed by a control policy plus a probing signal called an exploration. Then, we show input-to-state stability (ISS) and invariant admissibility of the closed-loop systems with the policies generated by integral policy iteration (I-PI) or invariantly admissible PI (IA-PI) method. Based on these, three online I-RL algorithms named explorized I-PI and integral Q -learning I, II are proposed, all of which generate the same convergent sequences as I-PI and IA-PI under the required excitation condition on the exploration. All the proposed methods are partially or completely model free, and can simultaneously explore the state space in a stable manner during the online learning processes. ISS, invariant admissibility, and convergence properties of the proposed methods are also investigated, and related with these, we show the design principles of the exploration for safe learning. Neural-network-based implementation methods for the proposed schemes are also presented in this paper. Finally, several numerical simulations are carried out to verify the effectiveness of the proposed methods.

Reinforcement learning interfaces for biomedical database systems.

PubMed

Rudowsky, I; Kulyba, O; Kunin, M; Parsons, S; Raphan, T

2006-01-01

Studies of neural function that are carried out in different laboratories and that address different questions use a wide range of descriptors for data storage, depending on the laboratory and the individuals that input the data. A common approach to describe non-textual data that are referenced through a relational database is to use metadata descriptors. We have recently designed such a prototype system, but to maintain efficiency and a manageable metadata table, free formatted fields were designed as table entries. The database interface application utilizes an intelligent agent to improve integrity of operation. The purpose of this study was to investigate how reinforcement learning algorithms can assist the user in interacting with the database interface application that has been developed to improve the performance of the system.

An Evaluation of Pedagogical Tutorial Tactics for a Natural Language Tutoring System: A Reinforcement Learning Approach

ERIC Educational Resources Information Center

Chi, Min; VanLehn, Kurt; Litman, Diane; Jordan, Pamela

2011-01-01

Pedagogical strategies are policies for a tutor to decide the next action when there are multiple actions available. When the content is controlled to be the same across experimental conditions, there has been little evidence that tutorial decisions have an impact on students' learning. In this paper, we applied Reinforcement Learning (RL) to…

Memory Transformation Enhances Reinforcement Learning in Dynamic Environments.

PubMed

Santoro, Adam; Frankland, Paul W; Richards, Blake A

2016-11-30

Over the course of systems consolidation, there is a switch from a reliance on detailed episodic memories to generalized schematic memories. This switch is sometimes referred to as "memory transformation." Here we demonstrate a previously unappreciated benefit of memory transformation, namely, its ability to enhance reinforcement learning in a dynamic environment. We developed a neural network that is trained to find rewards in a foraging task where reward locations are continuously changing. The network can use memories for specific locations (episodic memories) and statistical patterns of locations (schematic memories) to guide its search. We find that switching from an episodic to a schematic strategy over time leads to enhanced performance due to the tendency for the reward location to be highly correlated with itself in the short-term, but regress to a stable distribution in the long-term. We also show that the statistics of the environment determine the optimal utilization of both types of memory. Our work recasts the theoretical question of why memory transformation occurs, shifting the focus from the avoidance of memory interference toward the enhancement of reinforcement learning across multiple timescales. As time passes, memories transform from a highly detailed state to a more gist-like state, in a process called "memory transformation." Theories of memory transformation speak to its advantages in terms of reducing memory interference, increasing memory robustness, and building models of the environment. However, the role of memory transformation from the perspective of an agent that continuously acts and receives reward in its environment is not well explored. In this work, we demonstrate a view of memory transformation that defines it as a way of optimizing behavior across multiple timescales. Copyright © 2016 the authors 0270-6474/16/3612228-15$15.00/0.

Theoretical assumptions of Maffesoli's sensitivity and Problem-Based Learning in Nursing Education1

PubMed Central

Rodríguez-Borrego, María-Aurora; Nitschke, Rosane Gonçalves; do Prado, Marta Lenise; Martini, Jussara Gue; Guerra-Martín, María-Dolores; González-Galán, Carmen

2014-01-01

Objective understand the everyday and the imaginary of Nursing students in their knowledge socialization process through the Problem-Based Learning (PBL) strategy. Method Action Research, involving 86 students from the second year of an undergraduate Nursing program in Spain. A Critical Incident Questionnaire and Group interview were used. Thematic/categorical analysis, triangulation of researchers, subjects and techniques. Results the students signal the need to have a view from within, reinforcing the criticism against the schematic dualism; PBL allows one to learn how to be with the other, with his mechanical and organic solidarity; the feeling together, with its emphasis on learning to work in group and wanting to be close to the person taking care. Conclusions The great contradictions the protagonists of the process, that is, the students experience seem to express that group learning is not a form of gaining knowledge, as it makes them lose time to study. The daily, the execution time and the imaginary of how learning should be do not seem to have an intersection point in the use of Problem-Based Learning. The importance of focusing on the daily and the imaginary should be reinforced when we consider nursing education. PMID:25029064

Extinction of Pavlovian conditioning: The influence of trial number and reinforcement history.

PubMed

Chan, C K J; Harris, Justin A

2017-08-01

Pavlovian conditioning is sensitive to the temporal relationship between the conditioned stimulus (CS) and the unconditioned stimulus (US). This has motivated models that describe learning as a process that continuously updates associative strength during the trial or specifically encodes the CS-US interval. These models predict that extinction of responding is also continuous, such that response loss is proportional to the cumulative duration of exposure to the CS without the US. We review evidence showing that this prediction is incorrect, and that extinction is trial-based rather than time-based. We also present two experiments that test the importance of trials versus time on the Partial Reinforcement Extinction Effect (PREE), in which responding extinguishes more slowly for a CS that was inconsistently reinforced with the US than for a consistently reinforced one. We show that increasing the number of extinction trials of the partially reinforced CS, relative to the consistently reinforced CS, overcomes the PREE. However, increasing the duration of extinction trials by the same amount does not overcome the PREE. We conclude that animals learn about the likelihood of the US per trial during conditioning, and learn trial-by-trial about the absence of the US during extinction. Moreover, what they learn about the likelihood of the US during conditioning affects how sensitive they are to the absence of the US during extinction. Copyright © 2017 Elsevier B.V. All rights reserved.

Reinforcement Learning for Weakly-Coupled MDPs and an Application to Planetary Rover Control

NASA Technical Reports Server (NTRS)

Bernstein, Daniel S.; Zilberstein, Shlomo

2003-01-01

Weakly-coupled Markov decision processes can be decomposed into subprocesses that interact only through a small set of bottleneck states. We study a hierarchical reinforcement learning algorithm designed to take advantage of this particular type of decomposability. To test our algorithm, we use a decision-making problem faced by autonomous planetary rovers. In this problem, a Mars rover must decide which activities to perform and when to traverse between science sites in order to make the best use of its limited resources. In our experiments, the hierarchical algorithm performs better than Q-learning in the early stages of learning, but unlike Q-learning it converges to a suboptimal policy. This suggests that it may be advantageous to use the hierarchical algorithm when training time is limited.

Cognitive Components Underpinning the Development of Model-Based Learning

PubMed Central

Potter, Tracey C.S.; Bryce, Nessa V.; Hartley, Catherine A.

2016-01-01

Reinforcement learning theory distinguishes “model-free” learning, which fosters reflexive repetition of previously rewarded actions, from “model-based” learning, which recruits a mental model of the environment to flexibly select goal-directed actions. Whereas model-free learning is evident across development, recruitment of model-based learning appears to increase with age. However, the cognitive processes underlying the development of model-based learning remain poorly characterized. Here, we examined whether age-related differences in cognitive processes underlying the construction and flexible recruitment of mental models predict developmental increases in model-based choice. In a cohort of participants aged 9–25, we examined whether the abilities to infer sequential regularities in the environment (“statistical learning”), maintain information in an active state (“working memory”) and integrate distant concepts to solve problems (“fluid reasoning”) predicted age-related improvements in model-based choice. We found that age-related improvements in statistical learning performance did not mediate the relationship between age and model-based choice. Ceiling performance on our working memory assay prevented examination of its contribution to model-based learning. However, age-related improvements in fluid reasoning statistically mediated the developmental increase in the recruitment of a model-based strategy. These findings suggest that gradual development of fluid reasoning may be a critical component process underlying the emergence of model-based learning. PMID:27825732

Closed-loop adaptation of neurofeedback based on mental effort facilitates reinforcement learning of brain self-regulation.

PubMed

Bauer, Robert; Fels, Meike; Royter, Vladislav; Raco, Valerio; Gharabaghi, Alireza

2016-09-01

Considering self-rated mental effort during neurofeedback may improve training of brain self-regulation. Twenty-one healthy, right-handed subjects performed kinesthetic motor imagery of opening their left hand, while threshold-based classification of beta-band desynchronization resulted in proprioceptive robotic feedback. The experiment consisted of two blocks in a cross-over design. The participants rated their perceived mental effort nine times per block. In the adaptive block, the threshold was adjusted on the basis of these ratings whereas adjustments were carried out at random in the other block. Electroencephalography was used to examine the cortical activation patterns during the training sessions. The perceived mental effort was correlated with the difficulty threshold of neurofeedback training. Adaptive threshold-setting reduced mental effort and increased the classification accuracy and positive predictive value. This was paralleled by an inter-hemispheric cortical activation pattern in low frequency bands connecting the right frontal and left parietal areas. Optimal balance of mental effort was achieved at thresholds significantly higher than maximum classification accuracy. Rating of mental effort is a feasible approach for effective threshold-adaptation during neurofeedback training. Closed-loop adaptation of the neurofeedback difficulty level facilitates reinforcement learning of brain self-regulation. Copyright © 2016 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.

Ecological Momentary Assessment of Negative Symptoms in Schizophrenia: Relationships to Effort Based Decision Making and Reinforcement Learning

PubMed Central

Moran, Erin K.; Culbreth, Adam J.; Barch, Deanna M.

2017-01-01

Negative symptoms are a core clinical feature of schizophrenia, but conceptual and methodological problems with current instruments can make their assessment challenging. One hypothesis is that current symptom assessments may be influenced by impairments in memory and may not be fully reflective of actual functioning outside of the laboratory. The present study sought to investigate the validity of assessing negative symptoms using ecological momentary assessment (EMA). Participants with schizophrenia (N=31) completed electronic questionnaires on smartphones four times a day for one week. Participants also completed Effort-Based Decision Making and Reinforcement Learning (RL) tasks to assess the relationship between EMA and laboratory measures, which tap into negative symptom relevant domains. Hierarchical linear modeling analyses revealed that clinician-rated and self-report measures of negative symptoms were significantly related to negative symptoms assessed via EMA. However, working memory moderated the relationship between EMA and retrospective measures of negative symptoms, such that there was a stronger relationship between EMA and retrospective negative symptom measures among individuals with better working memory. We also found that negative symptoms assessed via EMA were related to poor performance on the Effort task, while clinician-rated symptoms and self-reports were not. Further, we found that negative symptoms were related to poorer performance on learning reward contingencies. Our findings suggest that negative symptoms can be assessed through EMA and that working memory impairments frequently seen in schizophrenia may affect recall of symptoms. Moreover, these findings suggest the importance of examining the relationship between laboratory tasks and symptoms assessed during daily life. PMID:27893230

Machine Learning Control For Highly Reconfigurable High-Order Systems

DTIC Science & Technology

2015-01-02

develop and flight test a Reinforcement Learning based approach for autonomous tracking of ground targets using a fixed wing Unmanned...Reinforcement Learning - based algorithms are developed for learning agents’ time dependent dynamics while also learning to control them. Three algorithms...to a wide range of engineering- based problems . Implementation of these solutions, however, is often complicated by the hysteretic, non-linear,

Fear of losing money? Aversive conditioning with secondary reinforcers.

PubMed