ABSTRACT

Although several conventional methods (e.g., pruning techniques) exist that may be used to automatically determine neural network con€guration and weights, they are often susceptible to trapping at local optima and characteristically dependent on the initial network structure (Gao et al., 1999). Overcoming these limitations would thus prove useful in the development of more intelligent neural network systems, those capable of demonstrating effective global search characteristics with fast convergence, and capable of providing alternative tools for modeling complex natural processes. The movement toward more intelligent network systems requires consideration of alternative strategies to help optimize neural network structure and aid in bringing multifaceted problems into focus. By converging on speci€c methodologies common to, for example, genetic algorithms, evolutionary programming, and fuzzy logic, we can begin to comprehend how when combined with neural networks, hybrid technology can impart the ef€ciency and accuracy needed in fundamental research, where multidisciplinary and multiobjective tasks are routinely performed. The objectives of this chapter are twofold: (1) to present theoretical concepts behind unconventional methodologies and (2) showcase a variety of example hybrid techniques, including neuro-fuzzy, neuro-genetic, and neuro-fuzzy-genetic systems.