Classifying Patterns in a Spiking Neural Network

Learning rules for spiking neural networks have emerged that can classify spatio-temporal spiking patterns as precise target spike trains, although there remains uncertainty in which rule to select that offers the greatest performance. Here, we quantify the performance of a stochastic neuron model in learning to classify input patterns by precise target responses as outputs, and compare its performance against other learning rules. We achieve a level of performance that is comparable with that found previously for alternative neuron models, and demonstrate the advantages of classifying inputs by multiple-spike timings: both by increasing the performance and the reliability of classifications.