DEDICATED TO MY FATHER and MOTHER Who taught me many useful algorithms. PREFACE Learning constitutes one of the most important phase of the whole psychological processes and it is essential in many ways for the occurrence of necessary changes in the behavior of adjusting organisms. In a broad sense influence of prior behavior and its consequence upon subsequent behavior is usually accepted as a definition of learning. Till recently learning was regarded as the prerogative of living beings. But

more »

... n the past few decades there have been attempts to construct learning machines or systems with considerable success. This book deals with a powerful class of learning algorithms that have been developed over the past two decades in the context of learning systems modelled by finite state probabilistic automaton. These algorithms are very simple iterative schemes. Mathematically these algorithms define two distinct classes of Markov processes with unit simplex (of suitable dimension) as its state space. The basic problem of learning is viewed as one of finding conditions on the algorithm such that the associated Markov process has prespecified asymptotic behavior. As a prerequisite a first course in analysis and stochastic processes would be an adequate preparation to pursue the development in various chapters. Automaton models and algorithms for learning, which is the central theme of the present book, have been widely accepted as one of the fundamental approaches to machine intelligence. In spite of this wide spread acceptability most of the published material on this topic to date are confined to articles in research journals or isolated chapters in anthologies and as such there is no one single text book or monograph on this topic. This book is intended to fill this gap. iv The book consists of two parts: Part I developes the theory of design of learning algorithms. In particular it deals with the convergence behavior of various classes of learning algorithms. Part II describes the application of these algorithms to decentralized decision making problems with incomplete information. More specifically, we deal with the applications to two person zero sum sequential games with incomplete information, two person team problem with incomplete information and Markov decision problem with incomplete information, Our aim is to provide a unified framework and a comprehensive treatment of most of the results pertaining to both the theory and applications known to date. Interestingly enough, much of the results presented in Part II came into existence only during the past three to four years.