Моделювання імпульсної нейронної мережі у задачі розпізнавання багатовимірних імпульсних послідовностей

Abstract

Problem of multidimensional pulse series recognition and possible ways of its solving were considered. For recognition problem solving pulsed neuron network, consisted of pulsed (or LIF — Leaky Integrate-and-Fire) neuron with recurrent connections was used. To determine the best algorithm by the criterion of validity and the error value, back propagation and linear classification algorithms were used for the network training. Analysis of the results testifies that the best algorithm is the linear classification one.