A New Evolution based Training Method for Dynamic Synapse Neural Networks

S. Saeb, M.B. Menhaj, S. Ali Seyyedsalehi (Iran)


Neural Networks, Dynamic Synapse, Synaptic Plasticity,Genetic Algorithm, Training, and Speech Recognition


A new evolution based approach is presented to regulate dynamic synapse neural network (DSNN) parameters. The DSNN is assigned to perform word recognition task on Persian digits. The wavelet packet transform is applied to the speech signal in order to decompose it into frequency sub-bands, before being introduced to the neural network. The new evolution based method is inspired by recombination accomplished by some insects such as bees. In this way, the recombination is done so that one chromosome is recombined with the best chromosome that exists in the present population. The simulations done on different sets of words indicate a better performance for the proposed method, from a training run-time point of view.

Important Links:

Go Back