Neural Network Weight Selection Using Genetic Algorithms
David J. Montana · 2002
Neural networks are a computational paradigm modeled on the human brain that has become popular in recent years for a few reasons. First, despite their simple structure, they provide very general computational capabilities [HORN89]. Second, they can be manufactured directly in VLSI hardware and hence provide the potential for relatively inexpensive massive parallelism [MEAD89]. Most importantly, they can adapt themselves to different tasks, i.e. learn, solely by selection of numerical “weights”. How to select these weights is a key issue in the use of neural networks. The usual approach is to derive a special-purpose weight selection algorithm for each neural network architecture. Here, we dicuss a different approach. Genetic algorithms are a class of search algorithms modeled on the process of natural evolution. They have been shown in practice to be very effective at function optimization, efficiently searching large and complex (multimodal, discontinuous, etc.) spaces to find nearly global optima. The search space associated with a neural network weight selection problem is just such a space. In this chapter, we investigate the utilization of genetic algorithms for neural network weight selection.