Self-organisation: a derivation from first principles of a class oflearning algorithms

摘要

reconstruction distortion measure.He includes in this distribution a contribution from the effect of codenoise which corrupts the output of the vector quantizer. Such code noisemodels the expected distorting effect of later stages of processing, andthus provides a convenient way of ensuring that the vector quantizeracquires a useful coding scheme. The neighborhood updating scheme ofKohonen's self-organizing neural network emerges as a special case ofthis code noise model. This reformulation of Kohonen's algorithmprovides a simple interpretation of the role of the neighborhood updatescheme which is used