COMPETITIVE LEARNING ALGORITHMS FOR ROBUST VECTOR QUANTIZATION

The efficient representation and encoding of signals with limited reso urces, e.g., finite storage capacity and restricted transmission bandw idth, is a fundamental problem in technical as well as biological info rmation processing systems. Typically, under realistic circumstances, the encoding and communication of messages has to deal with different sources of noise and disturbances. In this paper, we propose a unifyin g approach to data compression by robust vector quantization, which ex plicitly deals with channel noise, bandwidth limitations, and random e limination of prototypes. The resulting algorithm is able to limit the detrimental effect of noise in a very general communication scenario. In addition, the presented model allows us to derive a novel competit ive neural networks algorithm, which covers topology preserving featur e maps, the so-called neural-gas algorithm, and the maximum entropy so ft-max rule as special cases. Furthermore, continuation methods based on these noise models improve the codebook design by reducing the sens itivity to local minima. We show an exemplary application of the novel robust vector quantization algorithm to image compression for a telec onferencing system.