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

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

Recent developments in the quantitative analysis of complex networks, based largely on graph theory, have been rapidly translated to studies of brain network organization. The brain's structural and functional systems have features of complex networks--such as small-world topology, highly connected hubs and modularity--both at the whole-brain scale of human neuroimaging and at a cellular scale in non-human animals. In this article, we review studies investigating complex brain networks in diverse experimental modalities (including structural and functional MRI, diffusion tensor imaging, magnetoencephalography and electroencephalography in humans) and provide an accessible introduction to the basic principles of graph theory. We also highlight some of the technical challenges and key questions to be addressed by future developments in this rapidly moving field.

Complex brain networks: graph theoretical analysis of structural and functional systems

The authors present a unified account of 2 neural systems concerned with the development and expression of adaptive behaviors: a mesencephalic dopamine system for reinforcement learning and a “generic” error-processing system associated with the anterior cingulate cortex The existence of the error-processing system has been inferred from the error-related negativity (ERN), a component of the event-related brain potential elicited when human participants commit errors in reaction-time tasks The authors propose that the ERN is generated when a negative reinforcement learning signal is conveyed to the anterior cingulate cortex via the mesencephalic dopamine system and that this signal is used by the anterior cingulate cortex to modify performance on the task at hand They provide support for this proposal using both computational modeling and psychophysiological experimentation

/pdf/the-neural-basis-of-human-error-processing-reinforcement-246y7vy4ka.pdf

The neural basis of human error processing: Reinforcement learning, dopamine, and the error-related negativity.

The Reorienting System of the Human Brain: From Environment to Theory of Mind

The basal ganglia: a vertebrate solution to the selection problem?

We present a biologically plausible model of processing intrinsic to the basal ganglia based on the computational premise that action selection is a primary role of these central brain structures. By encoding the propensity for selecting a given action in a scalar value (the salience), it is shown that action selection may be re-cast in terms of signal selection. The generic properties of signal selection are defined and neural networks for this type of computation examined. A comparison between these networks and basal ganglia anatomy leads to a novel functional decomposition of the basal ganglia architecture into `selection' and `control' pathways. The former pathway performs the selection per se via a feedforward off-centre on-surround network. The control pathway regulates the action of the selection pathway to ensure its effective operation, and synergistically complements its dopaminergic modulation. The model contrasts with the prevailing functional segregation of basal ganglia into `direct' and `indirect' pathways.

A computational model of action selection in the basal ganglia. I. A new functional anatomy.

Recently, it has been demonstrated that several complex systems may have simple graph-theoretic characterizations as so-called ‘small-world’ and ‘scale-free’ networks. These networks have also been applied to the gross neural connectivity between primate cortical areas and the nervous system of Caenorhabditis elegans. Here, we extend this work to a specific neural circuit of the vertebrate brain—the medial reticular formation (RF) of the brainstem—and, in doing so, we have made three key contributions. First, this work constitutes the first model (and quantitative review) of this important brain structure for over three decades. Second, we have developed the first graph-theoretic analysis of vertebrate brain connectivity at the neural network level. Third, we propose simple metrics to quantitatively assess the extent to which the networks studied are small-world or scale-free. We conclude that the medial RF is configured to create small-world (implying coherent rapid-processing capabilities), but not scale-free, type networks under assumptions which are amenable to quantitative measurement.

/pdf/the-brainstem-reticular-formation-is-a-small-world-not-scale-7u9idfymrw.pdf

The brainstem reticular formation is a small-world, not scale-free, network

Is the short-latency dopamine response too short to signal reward error?

In a companion paper a new functional architecture was proposed for the basal ganglia based on the premise that these brain structures play a central role in behavioural action selection. The current paper quantitatively describes the properties of the model using analysis and simulation. The decomposition of the basal ganglia into selection and control pathways is supported in several ways. First, several elegant features are exposed--capacity scaling, enhanced selectivity and synergistic dopamine modulation--which might be expected to exist in a well designed action selection mechanism. The discovery of these features also lends support to the computational premise of selection that underpins our model. Second, good matches between model globus pallidus external segment output and globus pallidus internal segment and substantia nigra reticulata area output, and neurophysiological data, have been found which are indicative of common architectural features in the model and biological basal ganglia. Third, the behaviour of the model as a signal selection mechanism has parallels with some kinds of action selection observed in animals under various levels of dopaminergic modulation.

Tony J. Prescott

Papers

The basal ganglia: a vertebrate solution to the selection problem?

A computational model of action selection in the basal ganglia. I. A new functional anatomy.

The brainstem reticular formation is a small-world, not scale-free, network

Is the short-latency dopamine response too short to signal reward error?

A computational model of action selection in the basal ganglia. II. Analysis and simulation of behaviour.