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

I have developed "tennis elbow" from lugging this book around the past four weeks, but it is worth the pain, the effort, and the aspirin. It is also worth the (relatively speaking) bargain price. Including appendixes, this book contains 894 pages of text. The entire panorama of the neural sciences is surveyed and examined, and it is comprehensive in its scope, from genomes to social behaviors. The editors explicitly state that the book is designed as "an introductory text for students of biology, behavior, and medicine," but it is hard to imagine any audience, interested in any fragment of neuroscience at any level of sophistication, that would not enjoy this book. The editors have done a masterful job of weaving together the biologic, the behavioral, and the clinical sciences into a single tapestry in which everyone from the molecular biologist to the practicing psychiatrist can find and appreciate his or

Principles of Neural Science

A free-energy principle has been proposed recently that accounts for action, perception and learning. This Review looks at some key brain theories in the biological (for example, neural Darwinism) and physical (for example, information theory and optimal control theory) sciences from the free-energy perspective. Crucially, one key theme runs through each of these theories — optimization. Furthermore, if we look closely at what is optimized, the same quantity keeps emerging, namely value (expected reward, expected utility) or its complement, surprise (prediction error, expected cost). This is the quantity that is optimized under the free-energy principle, which suggests that several global brain theories might be unified within a free-energy framework.

/pdf/the-free-energy-principle-a-unified-brain-theory-3vz1twg7nv.pdf

The free-energy principle: a unified brain theory?

Brains, it has recently been argued, are essentially prediction machines. They are bundles of cells that support perception and action by constantly attempting to match incoming sensory inputs with top-down expectations or predictions. This is achieved using a hierarchical generative model that aims to minimize prediction error within a bidirectional cascade of cortical processing. Such accounts offer a unifying model of perception and action, illuminate the functional role of attention, and may neatly capture the special contribution of cortical processing to adaptive success. This target article critically examines this "hierarchical prediction machine" approach, concluding that it offers the best clue yet to the shape of a unified science of mind and action. Sections 1 and 2 lay out the key elements and implications of the approach. Section 3 explores a variety of pitfalls and challenges, spanning the evidential, the methodological, and the more properly conceptual. The paper ends (sections 4 and 5) by asking how such approaches might impact our more general vision of mind, experience, and agency.

/pdf/whatever-next-predictive-brains-situated-agents-and-the-epj56cs868.pdf

Whatever next? Predictive brains, situated agents, and the future of cognitive science

Uncertainty is critical in the measure of information and in assessing the accuracy of predictions. It is determined by probability P, being maximal at P = 0.5 and decreasing at higher and lower probabilities. Using distinct stimuli to indicate the probability of reward, we found that the phasic activation of dopamine neurons varied monotonically across the full range of probabilities, supporting past claims that this response codes the discrepancy between predicted and actual reward. In contrast, a previously unobserved response covaried with uncertainty and consisted of a gradual increase in activity until the potential time of reward. The coding of uncertainty suggests a possible role for dopamine signals in attention-based learning and risk-taking behavior.

/pdf/discrete-coding-of-reward-probability-and-uncertainty-by-2vdhyeux9s.pdf

Discrete coding of reward probability and uncertainty by dopamine neurons.

It is important for animals to estimate the value of rewards as accurately as possible. Because the number of potential reward values is very large, it is necessary that the brain's limited resources be allocated so as to discriminate better among more likely reward outcomes at the expense of less likely outcomes. We found that midbrain dopamine neurons rapidly adapted to the information provided by reward-predicting stimuli. Responses shifted relative to the expected reward value, and the gain adjusted to the variance of reward value. In this way, dopamine neurons maintained their reward sensitivity over a large range of reward values.

/pdf/adaptive-coding-of-reward-value-by-dopamine-neurons-3ewx08e9bf.pdf

Adaptive Coding of Reward Value by Dopamine Neurons

In this article, the authors report that the temporal precision of the reward prediction error signal encoded by midbrain dopamine neurons declines as the delay between a reward predicting stimulus and an expected reward increases. The temporal precision of the neural signal is qualitatively similar to that of anticipatory behavior.

The temporal precision of reward prediction in dopamine neurons

Rapid information transfer within the brain depends on chemical signalling between neurons that is mediated primarily by glutamate and GABA (γ-aminobutyric acid), acting at ionotropic receptors to cause excitatory or inhibitory postsynaptic potentials (EPSPs or IPSPs), respectively. In addition, synaptically released glutamate acts on metabotropic receptors to excite neurons on a slower timescale through second-messenger cascades, including phosphoinositide hydrolysis1. We now report a unique IPSP mediated by the activation of metabotropic glutamate receptors. In ventral midbrain dopamine neurons, activation of metabotropic glutamate receptors (mGluR1) mobilized calcium from caffeine/ryanodine-sensitive stores and increased an apamin-sensitive potassium conductance. The underlying potassium conductance and dependence on calcium stores set this IPSP apart from the slow IPSPs described so far2,3,4. The mGluR-induced hyperpolarization was dependent on brief exposure to agonist, because prolonged application of exogenous agonist desensitized the hyperpolarization and caused the more commonly reported depolarization1,5,6. The rapid rise and brief duration of synaptically released glutamate in the extracellular space can therefore mediate a rapid excitation through activation of ionotropic receptors, followed by inhibition through the mGluR1 receptor. Thus the idea that glutamate is solely an excitatory neurotransmitter must be replaced with a more complex view of its dual function in synaptic transmission.

Glutamate mediates an inhibitory postsynaptic potential in dopamine neurons

The acknowledged importance of uncertainty in economic decision making has stimulated the search for neural signals that could influence learning and inform decision mechanisms. Current views distinguish two forms of uncertainty, namely risk and ambiguity, depending on whether the probability distributions of outcomes are known or unknown. Behavioural neurophysiological studies on dopamine neurons revealed a risk signal, which covaried with the standard deviation or variance of the magnitude of juice rewards and occurred separately from reward value coding. Human imaging studies identified similarly distinct risk signals for monetary rewards in the striatum and orbitofrontal cortex (OFC), thus fulfilling a requirement for the mean variance approach of economic decision theory. The orbitofrontal risk signal covaried with individual risk attitudes, possibly explaining individual differences in risk perception and risky decision making. Ambiguous gambles with incomplete probabilistic information induced stronger brain signals than risky gambles in OFC and amygdala, suggesting that the brain's reward system signals the partial lack of information. The brain can use the uncertainty signals to assess the uncertainty of rewards, influence learning, modulate the value of uncertain rewards and make appropriate behavioural choices between only partly known options.

/pdf/explicit-neural-signals-reflecting-reward-uncertainty-2xmy4j85kt.pdf

Christopher D. Fiorillo

Papers

Discrete coding of reward probability and uncertainty by dopamine neurons.

Adaptive Coding of Reward Value by Dopamine Neurons

The temporal precision of reward prediction in dopamine neurons

Glutamate mediates an inhibitory postsynaptic potential in dopamine neurons

Explicit neural signals reflecting reward uncertainty