Search for optimal measure for discriminating spike trains with different randomness

Abstract

We wish to discriminate spike sequences based on the degree of irregularity. For this purpose, we search for a rational expressions of quadratic functions of consecutive interspike intervals that efficiently measures spiking irregularity. Under natural assumptions, the functional form of the coefficient can be parameterized by a single parameter. The parameter is determined so as to maximize the mutual information between the distributions of coefficients computed for spike sequences derived from different renewal point processes. We find that the local variation of interspike intervals, LV (Neural Comput. Vol. 15, pp. 2823-42, 2003), is nearly optimal for whose intrinsic irregularity is close to that of experimental data.