An Experience Based Competitive Learning Neural Model for Data Compression

Abstract

This paper presents a novel design of a multi-layer neural model which implements an experience based learning algorithm to train the neural network and achieve comparable data compression. This design takes 16 bits each cycle as the input to the network. A competitive learning based operation is then applied to the input to find the winner which exactly matches the 16 bits. Another hidden layer is further designed to produce various outputs corresponding to the different outcomes of the competitive learning. The output then controls the coder to complete the encoding. Finally, an experience based learning algorithm is developed to train the network to make the best use of the statistical information from input to achieve the highest possible compression.