Browse Books

Reviewer: Grigore Albeanu

To use heuristic approaches in problem solving, one needs a good understanding of the problem to be solved. Particular aspects of the problem should be used when designing a heuristic solver. This book represents the best guide-according to my knowledge-on designing heuristic methods for solving problems; it is a valuable contribution to the field. The material consists of three parts: "Fundamentals" (chapters 2 and 3), "Modern Heuristics" (chapters 4 through 6), and "Case Studies" (chapters 7 and 8). These parts-along with chapter 1, an introduction establishing the author's aim; chapter 9, an inspired summary presenting the learned lessons; a rich and relevant list of references; a list of notations; a glossary (to explain the abbreviations); and a valuable index-define the book's structure. In the following review, I'll present some of the book's highlights. The problems to be solved are mainly considered as optimization problems having some properties: difficulty, locality, and decomposability. The following measures and techniques are described in this context in chapter 2: asymptotic notations, complexity classes, the fitness-distance correlation coefficient, the autocorrelation function, polynomial decomposition, Walsh decomposition, and schemata analysis. In chapter 3, after a concise presentation of the standard optimization methods (analytical and numerical differentiation based methods, the simplex method, interior point methods, integer programming, uninformed and informed search, branch and bound, dynamic programming, and cutting plane methods), the author returns to the heuristic optimization methods. First, he analyzes basic aspects of heuristics methods and approximation algorithms. Considering modern heuristics (also called "meta-heuristics"), he presents details of the no-free-lunch theorem regarding the performance of optimization algorithms. Chapter 4 presents the main elements used in the design of modern heuristics: the representation, the search operator, the fitness function, and the initialization. Algorithms for both local search methods (variable neighborhood, "guided local search, iterated local search, simulating annealing, tabu search, and evolution strategies) and recombination-based search methods ([with a deep view on] genetic algorithms, distribution algorithms, and genetic programming)" are described in chapter 5. Locality and bias are two general principles for the design of modern heuristics. To solve optimization problems with high locality that must be preserved, "similarities between phenotypes must correspond to similarities between genotypes." Moreover, biasing representations, problem-specific search operators, initial solutions, fitness functions, or search strategies must exploit knowledge of the problem to be solved. These principles are presented in chapter 6 and used in Part 3's two case studies: grammatical evolution (chapter 7) and the optimal communication spanning tree (chapter 8). The case studies include experimental results and show the performance of the proposed approaches. I recommend this book to graduate students (no exercises are included in the book, but the content and the two case studies are appropriate teaching materials); practitioners (as a guide to choose between using standard approaches or designing new algorithms based on the principles described); and researchers (as a good reference). Online Computing Reviews Service