Stochastic approximation algorithms and applications (Q4346705)

This book is concerned with the qualitative and asymptotic properties of stochastic approximation algorithms in many diverse forms in which they arise in applications. The first three chapters deal with the classical Robbins-Monro and Kiefer-Wolfowitz methods and their modifications motivated by practical considerations. Many examples of applications to learning, state dependent noise, queueing as well as in signal processing and adaptive control are described. The five following chapters give the mathematical background concerning the convergence theorems of the algorithms (convergence in probability as well as weak convergence). The proofs of convergence and investigation of the rate of convergence of the applications mentioned before follow. The problem of selecting a good step size sequence for the algorithms is dealt with in the last but one chapter. The closing chapter concerns the decentralized and asynchronous forms of the stochastic approximation algorithms.