Automatic estimation of learning rate for LVQ networks

Abstract

Various learning vector quantization (LVQ) algorithms proposed before have made use of the learning coefficient alpha (t) for obtaining convergence of training. The authors present a new method that can be to estimate the learning rate at alpha (t) from the input and codebook vectors directly and independent of the past history of alpha (t). The results of experiments done with real speech data with the algorithm are also reported.<>