Spike pattern classification is a key topic in machine 27 learning, computational neuroscience and electronic 28 device design. Here we offer a new supervised learn29 ing rule based on Support Vector Machines to deter30 mine the synaptic weights of a leaky integrate-and31 fire neuron model for spike pattern classification. We 32 compare classification performance between this al33 gorithm and other methods sharing the same concep34 tual framework. We consider the effect of postsynap35 tic potential kernel dynamics on patterns separability, 36 and we propose an extension of the method to de37 crease computational load. The algorithm performs 38 well in generalization tasks. We show that the peak 39 value of spike patterns separability depends on a re40 lation between postsynaptic potential dynamics and 41 spike pattern duration, and we propose a particular 42 kernel that is well-suited for fast computations and 43 electronic implementations. © 2012 Ambard and Rotter.
Ambard, M., & Rotter, S. (2012). Support vector machines for spike pattern classification with a leaky integrate-and-fire neuron. Frontiers in Computational Neuroscience, (SEPTEMBER). https://doi.org/10.3389/fncom.2012.00078