D.N. LAPSHIN, A.A. LUKYANITSA
NEURAL NETWORK MODEL OF SIGNAL EXTRACTION FROM A COMPLEX SENSORY FLOW CONTAINING SELF-GENERATED NOISE
All animals inevitably have to solve the problem of extracting biologically
important information which comes to their sensory organs intermixed with
self-generated signals corresponding to an animal's own behavior such as
locomotion. This problem arose and, probably, has been successfully solved at
the early evolutionary stages.
The aim of this study was to find the minimal level of complexity of artificial neural network still able to effectively extract the 'external signal' from the background of 'self-generated noise'. We have demon-strated that provided with the information about the movements of an 'animal' ('motor copy' or 'corollary discharge') as well as with the first and second derivatives of the input signal the network consisting of 10-20 units, preliminary trained, is able to extract external signals from the background of high-ampli-tude 'self-generated noise'.