Self-organized spiking neural network recognizing phase/frequency correlations

Training a spiking neural network usually implies formation of groups of cooperating neurons capable of recognizing various spatiotemporal structures in input signal. One of the most important structural features of signal is presence of episodes when frequency of pulses arriving from certain sets of inputs (receptors) or single pulse timing become correlated or partially synchronized. Recognition of signal synchrony under a wide and a priori unknown range of conditions still remains a difficult problem despite numerous network models proposed for its solution. In the present work this problem is solved by a network based on a neuron model being an extension of commonly used leaky integrate-and-fire neuron. High synchrony recognition reliability for signals with strongly varying parameters is achieved by this model due to introduced neuron stabilization mechanism, homeostatic properties of the network and proposed variant of synaptic plasticity laws which changes synaptic weights in two cases - when neuron fires and when postsynaptic pulse generation is suppressed by inhibitory synapses.