Using the Functional Behavior of Neurons for Genetic Recombination in Neural Nets Training

Abstrac t . We propose a new hyb rid genetic back propagation training algorithm based on a unique functional matching recombination method. T he method is used to evolve pop ulations of neur al networks and provides versatility in network architecture and activation functions. Net reorganization and reconstruction is carried out prior to genet ic recomb ina tion usin g a funct ional behavior correlation measure to compare the functional role of the var ious neurons. Compar ison is don e by correlating the intern al representations generated for a given training set. Net st ruc ture is dynamically changed during the evolutionary process, expanded by reorganization and reconstruct ion and trimmed by pruning unnecessary neur ons. The ab ility to change net structure throughout generations allows t he net population to fit itself to the requirements of dynamic adaptation , performance, and size considerations in the selection process , t hus generating smaller and mor e efficient nets t hat are likely to have higher generalization cap ab iliti es. A func t ional behavior corr elation meas ure is used exte nsively to explore and compare nets and neurons, and its ability is demonst rat ed by investi gat ing the results of genetic recombination. The vit ality of nets organized via the functional behavior correlat ion measure pr ior to genet ic recombinati on is demonstrated by statist ical results of computer simulat ions. The performance of t he proposed method and its generalization capabilit ies are demonstrated using Parity, Symmetry and handwrit ten digit recognit ion training tasks.