A Hybrid Evolutionary Learning System for Synthesizing Neural Network Pattern Recognition Systems

摘要

An approach is introduced for developing neural network patterk pattern recognition systems using a hybrid evolutionary learning system for pattern recognition (HELPR) concept. A genetic algorithm is used to assemble detectors and pattern recognition systems while traditional weight training methods are used to determine welthts. The results show that this novel approach develops simpler neural topologies than cascade correlation and can do so using very simple training metrics.