Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives when interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the key ideas and algorithms of reinforcement learning. Their discussion ranges from the history of the field's intellectual foundations to the most recent developments and applications. The only necessary mathematical background is familiarity with elementary concepts of probability. The book is divided into three parts. Part I defines the reinforcement learning problem in terms of Markov decision processes. Part II provides basic solution methods: dynamic programming, Monte Carlo methods, and temporal-difference learning. Part III presents a unified view of the solution methods and incorporates artificial neural networks, eligibility traces, and planning; the two final chapters present case studies and consider the future of reinforcement learning.

/pdf/reinforcement-learning-an-introduction-rzxgej9p17.pdf

Reinforcement Learning: An Introduction

抗原变异可使得多种致病微生物易于逃避宿主免疫应答。表达在感染红细胞表面的恶性疟原虫红细胞表面蛋白1（PfPMP1）与感染红细胞、内皮细胞、树突状细胞以及胎盘的单个或多个受体作用，在黏附及免疫逃避中起关键的作用。每个单倍体基因组var基因家族编码约60种成员，通过启动转录不同的var基因变异体为抗原变异提供了分子基础。

疟原虫var基因转换速率变化导致抗原变异[英]／Paul H, Robert P, Christodoulou Z, et al//Proc Natl Acad Sci U S A

Machine Learning is the study of methods for programming computers to learn. Computers are applied to a wide range of tasks, and for most of these it is relatively easy for programmers to design and implement the necessary software. However, there are many tasks for which this is difficult or impossible. These can be divided into four general categories. First, there are problems for which there exist no human experts. For example, in modern automated manufacturing facilities, there is a need to predict machine failures before they occur by analyzing sensor readings. Because the machines are new, there are no human experts who can be interviewed by a programmer to provide the knowledge necessary to build a computer system. A machine learning system can study recorded data and subsequent machine failures and learn prediction rules. Second, there are problems where human experts exist, but where they are unable to explain their expertise. This is the case in many perceptual tasks, such as speech recognition, hand-writing recognition, and natural language understanding. Virtually all humans exhibit expert-level abilities on these tasks, but none of them can describe the detailed steps that they follow as they perform them. Fortunately, humans can provide machines with examples of the inputs and correct outputs for these tasks, so machine learning algorithms can learn to map the inputs to the outputs. Third, there are problems where phenomena are changing rapidly. In finance, for example, people would like to predict the future behavior of the stock market, of consumer purchases, or of exchange rates. These behaviors change frequently, so that even if a programmer could construct a good predictive computer program, it would need to be rewritten frequently. A learning program can relieve the programmer of this burden by constantly modifying and tuning a set of learned prediction rules. Fourth, there are applications that need to be customized for each computer user separately. Consider, for example, a program to filter unwanted electronic mail messages. Different users will need different filters. It is unreasonable to expect each user to program his or her own rules, and it is infeasible to provide every user with a software engineer to keep the rules up-to-date. A machine learning system can learn which mail messages the user rejects and maintain the filtering rules automatically. Machine learning addresses many of the same research questions as the fields of statistics, data mining, and psychology, but with differences of emphasis. Statistics focuses on understanding the phenomena that have generated the data, often with the goal of testing different hypotheses about those phenomena. Data mining seeks to find patterns in the data that are understandable by people. Psychological studies of human learning aspire to understand the mechanisms underlying the various learning behaviors exhibited by people (concept learning, skill acquisition, strategy change, etc.).

Machine learning

▪ Abstract The prefrontal cortex has long been suspected to play an important role in cognitive control, in the ability to orchestrate thought and action in accordance with internal goals. Its neural basis, however, has remained a mystery. Here, we propose that cognitive control stems from the active maintenance of patterns of activity in the prefrontal cortex that represent goals and the means to achieve them. They provide bias signals to other brain structures whose net effect is to guide the flow of activity along neural pathways that establish the proper mappings between inputs, internal states, and outputs needed to perform a given task. We review neurophysiological, neurobiological, neuroimaging, and computational studies that support this theory and discuss its implications as well as further issues to be addressed

/pdf/an-integrative-theory-of-prefrontal-cortex-function-2j9unprn5f.pdf

An integrative theory of prefrontal cortex function

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

Hippocampal novelty responses studied with functional neuroimaging

Standard economic indicators provide an incomplete picture of what we value both as individuals and as a society. Furthermore, canonical macroeconomic measures, such as GDP, do not account for non-market activities (e.g., cooking, childcare) that nevertheless impact well-being. Here, we introduce a computational tool that measures the subjective reward value of experiences (e.g., playing a musical instrument without errors). We go on to validate this tool with neural data, using fMRI to measure neural activity in subjects performing a reinforcement learning task that incorporated periodic ratings of subjective affective state. Learning performance determined level of payment (i.e., extrinsic reward). Crucially, the task also incorporated a skilled performance component (i.e., intrinsic reward) which did not influence payment. Both extrinsic and intrinsic rewards influenced affective dynamics, and their relative influence could be captured in our computational model. Individuals for whom intrinsic rewards had a greater influence on affective state than extrinsic rewards had greater ventromedial prefrontal cortex (vmPFC) activity for intrinsic than extrinsic rewards. Thus, we show that computational modelling of affective dynamics can index the subjective value of intrinsic relative to extrinsic rewards, a "computational hedonometer" that reflects both behavior and neural activity that quantifies the subjective reward value of experience.

/pdf/a-neurocomputational-model-for-intrinsic-reward-43ogj33mby.pdf

A neurocomputational model for intrinsic reward

Substantial evidence indicates that subjective value is adapted to the statistics of reward expected within a given temporal context. However, how these contextual expectations are learned is poorly understood. To examine such learning, we exploited a recent observation that participants performing a gambling task adjust their preferences as a function of context. We show that, in the absence of contextual cues providing reward information, an average reward expectation was learned from recent past experience. Learning dependent on contextual cues emerged when two contexts alternated at a fast rate, whereas both cue-independent and cue-dependent forms of learning were apparent when two contexts alternated at a slower rate. Motivated by these behavioral findings, we reanalyzed a previous fMRI data set to probe the neural substrates of learning contextual reward expectations. We observed a form of reward prediction error related to average reward such that, at option presentation, activity in ventral tegmental area/substantia nigra and ventral striatum correlated positively and negatively, respectively, with the actual and predicted value of options. Moreover, an inverse correlation between activity in ventral tegmental area/substantia nigra (but not striatum) and predicted option value was greater in participants showing enhanced choice adaptation to context. The findings help understanding the mechanisms underlying learning of contextual reward expectation.

/pdf/learning-contextual-reward-expectations-for-value-adaptation-4rqmwrb3tw.pdf

Learning Contextual Reward Expectations for Value Adaptation.

Theories of monoaminergic neurotransmitter dys regulation continue to provide the principal framework for the investigation of the pathophysiology of depression and schizophrenia (Losonczy et al, 1987; Willner et al, 1992). The availability of techniques enabling quantitative pharmacological measurements, such as receptor number, to be made in vivo has heightened expectations for an understanding of the mechanisms underlying these conditions. Dopamin ergic receptor systems have been a particular target for in-vivo studies. However, the conflicting results from research based upon these approaches have been a source of disappointment (Wong et al, 1986; Farde et a!, 1990). As little is known concerning the functional role of dopamine in higher brain function in humans, findings based upon quantitative pharmacological measurements are in themselves unlikely to provide a sufficient theoretical basis for a mechanistic understanding of psychiatric disease. A fundamental knowledge of normal brain function is almost certainly a prerequisite for the elaboration of theories of psychiatric disease. Higher brain functions are expressed primarily at a systems level that depend upon an interaction of large populations of neurons organised into complex, often anatomically distributed, systems (Edelman & Mountcastle, 1977). Dynamic interactions between these systems are subject to modifications by a multitude of factors that include neurotransmitter inputs (Edelman, 1987). Therefore a theory of psychiatric illness based upon a putative neuro transmitter dysfunction must address the regulation of neural processes subserving normal psychological functioning. Furthermore, it should explain how neurotransmitter dysregulation leads to the emergence of the psychological symptoms and pathophysiological dysfunctions that characterise psychiatric syndromes. Few, if any, current neurotransmitter theories of psychiatric illness meet these criteria. Undoubtedly, the limiting factor in relation to the monoaminergic-based theories is the paucity of neuroscientific information regarding their regulatory role on cortical function. Findings based upon work in non-human primates have now begun to establish the role of neurotransmitter systems in the regulation of the neural systems subserving specific psychological states (Goldman-Rakic, 1988). An important goal in research on humans is to establish similar links. Novel applications of functional imaging techniques can provide the necessary links by specifying the neuroanatoinical locus of interactions between neural systems subserving specific psychological processes and neurotransmitter inputs (Friston et a!, 199lb). The theoretical basis of such research, with a particular emphasis on dopamine neurotransmission, and an account of preliminary findings from studies in humans, using positron emission tomography (PET), are the basis of this paper.

/pdf/exploring-the-functional-role-of-monoaminergic-1kisnholc1.pdf

Raymond J. Dolan

Papers

Hippocampal novelty responses studied with functional neuroimaging

A neurocomputational model for intrinsic reward

Learning Contextual Reward Expectations for Value Adaptation.

Exploring the functional role of monoaminergic neurotransmission. A method for exploring neurotransmitter dysfunction in psychiatric disorders.

15. Functional imaging dissociations within right prefrontal cortex during episodic memory retrieval