Competitive neural network scheme for learning vector quantisation

Abstract

A novel self-development neural network scheme, which employs two resource counters to record node activity, is presented. The proposed network not only harmonises equi-error and equiprobable criteria, but it also avoids the stability-and-plasticity dilemma. Simulation results show that the new scheme displays superior performance (in terms of measured MSE, MAE, and training speed) over other neural network models.