Approach with Reducing Time-Delays to Rise Prediction Ability for Neural Networks

Abstract

The time-delays induce instability, then the instability influences the prediction accuracy of neural networks, hence, time-delays should be considered as a key indicator for constructing neural networks. Based on above mentioned problem, we reconstructed time-dependent Lyapunov function. Next following, we found the sufficient condition which ensured the convergence of neural networks and prevented highfrequency oscillation, which effectively increased the prediction accuracy of neural networks. This results show that the prediction accuracy about raise 2.5% for neural networks after reducing time-delays, and the mean square error decreased about 0.02 for training samples using neural networks.