Knowledge Incorporation in Evolutionary Computation [Book Review]

I n recent years, we have witnessed a rapidly growing research activity that incorporates learning techniques, human preferences, and domain knowledge into evolutionary algorithms for solving complex search and optimization problems. It reflects a basic principle that is theoretically supported by the No Free Lunch Theorem: an efficient and effective algorithm should be able to adapt itself to the problem structure and thus problem-specific. Although numerous research papers on incorporation of knowledge into evolutionary computation can be found in a range of academic journals and conferences, there is a lack of suitable reference books in this area for researchers and practitioners. This edited book aims at addressing this challenge. It carefully classifies the various research topics and presents the state-of-the-art of both techniques and applications in a systematic way. The book is divided into six parts. Part I of the book contains a single chapter written by Xin Yao that provides a concise yet insightful introduction to evolutionary computation. Apart from a generic framework for evolutionary algorithms, theoretical and practical aspects concerning the benefit of using a population, search step-size adaptation, and constraint handling are discussed. Six chapters are collected in Part II describing various approaches to the embedding of a priori knowledge and domain knowledge acquired from the previous search into population initial-ization, recombination, and mutation. Chapter 2 shows how knowledge stored in the form of memory can be helpful in genetic p r o g r a m m i n g-based symbol regression and robotic control. The application of cultural algorithms, a class of evolutionary methods that extract and reuse domain knowledge , to a job shop scheduling problem is presented in Chapter 3. Knowledge extraction and incorporation using case-based reasoning techniques is described in Chapter 4. A chained cultural genetic programming algorithm in which two populations communicate by means of a shared belief space is presented in Chapter 5. The effectiveness of this algorithm is demonstrated in solving nonlinear program problems and simulating consumer markets. The reduction of randomness in mutation and crossover is advocated in Chapter 6, where polyno-mials and neural networks, which are two of the most widely-used surrogate models that are able to exploit domain knowledge hidden in the previous search data and thus speed up evolution, are employed to access the quality of candidate solutions. Consequently, those fitter solutions according to the surrogates are chosen as the final offspring. Part II concludes …