Provides a general mathematical framework for the analytical aspects of stochastic automata. Shows that under certain topological conditions, non-deterministic automata are generated, which in some cases are produced by stochastic automata.

Markov Chains -- Direct Methods -- Iterative Methods -- Projection Methods -- Block Hessenberg Matrices -- Decompositional Methods -- LI-Cyclic Markov -- Chains -- Transient Solutions -- Stochastic Automata Networks -- Software.

Networks of Learning Automata: Techniques for Online Stochastic Optimization is a comprehensive account of learning automata models with emphasis on multiautomata systems. It considers synthesis of complex learning structures from simple building blocks and uses stochastic algorithms for refining probabilities of selecting actions. Mathematical analysis of the behavior of games and feedforward networks is provided. Algorithms considered here can be used for online optimization of systems based on noisy measurements of performance index. Also, algorithms that assure convergence to the global optimum are presented. Parallel operation of automata systems for improving speed of convergence is described. The authors also include extensive discussion of how learning automata solutions can be constructed in a variety of applications.

Ongoing advances in science and engineering enable mankind to design and operate increasingly sophisticated systems. Both their design and operation require the understanding of the system and its interaction with the envir- ment. This necessitates the formalisation of the knowledge about the system by models. A major issue is what kind of model is best suited for a given task. This book is about the supervision of continuous dynamical systems. Such systems are typically described by di?erential equations. However, this does notautomaticallymeanthatdi?erentialequationsarepropermodelsforso- ing supervision tasks. Instead, this book and recent approaches in literature show that supervision tasks do in general not require the use of such precise modelsasdi?erentialequations.Thisisofinterestbecauseuncertainties,t- ically occurring in supervision, make the use of precise models very di?cult. Alternative approaches therefore use less precise models such as discrete– event descriptions to solve supervision tasks on a higher level of abstraction. Discrete–event descriptions in form of automata are one of the key elements of this book. To reach this higher level of abstraction, uncertainties by qu- tisation are introduced on purpose, taking into account a loss of precision. This is one of the main di?erence to other approaches. When using nume- calmodelsliketransferfunctionsordi?erentialequations,uncertaintiesmake the analysis more di?cult. Not so here, where the system is described on a qualitative level on which uncertainties are naturally incorporated. The book presents a new way to describe systems for supervision. Preparing this book I learned that the key to solve supervision problems is simplicity.

Introduction to Probabilistic Automata deals with stochastic sequential machines, Markov chains, events, languages, acceptors, and applications. The book describes mathematical models of stochastic sequential machines (SSMs), stochastic input-output relations, and their representation by SSMs. The text also investigates decision problems and minimization-of-states problems arising from concepts of equivalence and coverings for SSMs. The book presents the theory of nonhomogeneous Markov chains and systems in mathematical terms, particularly in relation to asymptotic behavior, composition (direct sum or product), and decomposition. "Word functions," induced by Markov chains and valued Markov systems, involve characterization, equivalence, and representability by an underlying Markov chain or system. The text also discusses the closure properties of probabilistic languages, events and their relation to regular events, particularly with reference to definite, quasidefinite, and exclusive events. Probabilistic automata theory has applications in information theory, control, learning theory, pattern recognition, and time sharing in computer programming. Programmers, computer engineers, computer instructors, and students of computer science will find the collection highly valuable.

In recent years works done by most researchers towards building autonomous intelligent controllers frequently mention the need for a methodology of design and a measure of how successful the final result is. This monograph introduces a design methodology for intelligent controllers based on the analytic theory of intelligent machines introduced by Saridis in the 1970s. The methodology relies on the existing knowledge about designing the different sub-systems composing an intelligent machine. Its goal is to provide a performance measure applicable to any of the sub-systems, and use that measure to learn on-line the best among the set of pre-designed alternatives, given the state of the environment where the machine operates. Different designs can be compared using this novel approach.