Stochastic recursive algorithms for optimization. Simultaneous perturbation methods (Q441138): Difference between revisions

A general optimization problem considered in the present book has the form: Find \(\theta\in\mathbb{R}^n\) that minimizes \(J(\theta)\), \(\theta\in C\). If one has complete information about \(J\) and \(C\), this is a well-known deterministic optimization problem. If \(J\) is the expected value over noisy observations, then it is a stochastic optimization problem. Stochastic approximations have its roots in a famous paper by \textit{H. Robbins} and \textit{S. Monro} [Ann. Math. Stat. 22, 400--407 (1951; Zbl 0054.05901)]. \textit{J. Kiefer} and \textit{J. Wolfowitz} [Ann. Math. Stat. 23, 462--466 (1952; Zbl 0049.36601)] published the first stochastic approximation algorithm for stochastic optimization. The book under review summarizes the recent research on simultaneously perturbation problems. Many results of the text mainly based on the author's own contributions to this topic. The book provides a coverage of the known material in stochastic optimizations, such that both researchers and practitioners should find it useful. The text is well understandable, the book is clearly written and impressively printed. Theorems and algorithms are emphasized in coloured frames. Therefore, the book can be used as material for lectures dedicated to master students. The book is divided into 6 parts (14 chapters): 1. Introduction to stochastic recursive algorithms, -- 2. Gradient estimation schemes, -- 3. Hessian estimation schemes, -- 4. Variations to the basic scheme, -- 5. Applications, -- 6. Appendix. There are references at the end of every chapter. The appendix informs on martingales, ODE, the Borkar-Meyn theorem and the Kushner-Clark theorem. Applications are studied in service systems, road traffic control and in communication networks.