Statistical approach to unsupervised recognition of spatio-temporal patterns by spiking neurons

The problem of unsupervised recognition of spatio-temporal structure in the sensory signal by a network of spiking neurons is considered from the statistical point of view. A novel model of spiking neurons called SSN/st is proposed as a solution for this problem. Computational experiments with artificially generated data demonstrate that statistically sound unsupervised learning mechanism and generalized Hebbian laws of synaptic plasticity implemented in this model provide it with high sensitivity and robustness.