A model of a learning neural network for data compressing

Resources required by neuron net model and its performance depend greatly upon amount of receptor elements and hence upon number of different incoming signal (impulse) combinations. Input vector dimensions exceed computer performance capabilities. Because of this compression during data processing in first levels of neurosystems is a necessary. There are some well-known formal methods of compression, and methods based on the same principles are required for purposes of neurocomputing, to achieve conceptual and structure unification of the approach and the benefit of such traditional additions as noise-protection and ability to retune system during normal work (teach net). This article offers a realisation of a first level neural network model of an approach based on certain rules of creating neuron receptive fields. This net can adapt its structure and optimal number of elements to actual object environment. In other words structure and inner dimensions depend upon real spectre of incoming signals. Analysers of real living systems have analogous adaptation power. This was showed in numerous experiments with deprivation