A Hybrid Evolutionary System for Designing Artifical Neural Networks

This paper proposed a hybrid evolutionary system HPSONN to automatically design artificial neural networks (ANNpsilas), where ANNpsilas structure and parameters are tuned simultaneously. In HPSONN, an improved particle swarm optimization using optimal foraging theory (PSOOFT) and a binary particle swarm optimization (BPSO) are used to train ANNpsilas free parameters (weights and bias) and find optimal ANNpsilas structure, respectively. The experimental results on tool life prediction problem show that HPSONN can produce compact ANNs with good accuracy and generalization.