Competitive Neural Network Traing: A Multi-resolution Approach

摘要

A multi-resolution method for training a Kohonen competitive neural network (KCNN) is presented. Starting with a low resolution sample of the input data, the training algorithm is applied to a sequence of monotonically increasing-resolution samples of the given data. The final weight matrix obtained from a low resolution stage is used as the initial weight matrix for the next stage which is a higher resolution stage. In the average case the multi-resolution reduces the computation time by a factor of more than two with a slight improvement in the quality of quantization. Alternatively it can be used to identify two local optimum solutions at the same time the traditional KCNN finds one local optimum.