Customer Reviews

Introduction to Stochastic Search and Optimization

5 star:		(3)
4 star:		(0)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

Introduction to Stochastic Search and Optimization provides comprehensive, current information on methods for real-world problem solving, including stochastic gradient and non-gradient techniques, as well as relatively recent innovations such as simulated annealing, genetic algorithms, and MCMC. It is written to be read and understood by graduate students, industrial practitioners, and experienced researchers in the field. Web links to software and data sets, and an extensive list of references of the book allows the reader to explore deeper into certain topic areas. I also found the index to be very comprehensive and carefully done. The appendices are as a refresher and summary of much of the prerequisite material. The book is somewhat unique in providing a balanced discussion of algorithms, including both their strengths and weaknesses. The book is among very few books that have integrated essential parts of statistical fields with optimization and decision making. The book's inclusion of a chapter on optimal experimental design is an example of such integration. The approaches discussed in the book could be used for financial decision making, forecasting, and quality improvement, among many other areas.

I stumbled upon this book searching for a Genetic Algorithm book. The coverage of topics are unique and very interesting. This is the first book I came across that treats both the evolutionary algorithms (GA) and the stochastic search methods. Recursive Linear Estimator (e.g. Kalman Filter), Markov Chain Monte Carlo (e.g. Metropolis-Hastings, Gibbs), and Reinforcement Learning, are some of the stochastic material discussed. Continuous and discrete parameters are treated as well as noisy data, but not so much on constrained optimization.

The algorithms presented are very practical and theoretically well founded. When I learned about SPSA, I was most impressed to find out that it is possible to estimate the gradient by just two objective function calls (instead of finite differencing every dimension of the parameter vector to be optimized), and this is regardless of the number of dimensions of the parameter vector!

The book is aimed at rather general audiences in science and engineering. Rigorous mathematical details are avoided.

A must have for anyone interested in otimization! Extremely well written and objective.