‘Winner-take-all’ circuit for neurocomputing applications

Abstract

In the paper a parallel analogue implementation of the ‘winner-take-all’ function is proposed. It can be utilised in the hardware realisation of neurocomputer systems based on competitive learning, which find application in areas such as pattern classification, vector quantisation and image recognition. Analytic results are derived by means of the piecewise-linear method in the case of two variables and SPICE simulation results are presented in the general case. It is shown that the circuit exhibits some ad hoc properties that make it well suited for this kind of application.