Estimating the similarity of temporal discharge patterns of auditory neurons

Abstract

Behavioral data obtained from stimulus generalization experiments are often interpreted to imply how similar animal perceptions of test stimuli are to a training stimulus. An appropriate decision variable would be useful for evaluating physiological responses in the context of stimulus generalization tasks. The decision variable P(A) obtained from ROC analysis of spike count distributions is useful for evaluating rate coding schemes [e.g., Relkin and Pelli, J. Acoust. Soc. Am. 82, 1679–1691 (1987)], but it ignores temporal information which may exist in responses to periodic sounds. Neural autocorrelograms show temporal firing patterns following a spike and were generated for cochlear nucleus unit responses to aperiodic, quasiperiodic, and periodic stimuli. Variance from average rate was computed as a function of time for each autocorrelogram; similarity between two autocorrelograms was estimated as the ratio of these variances. A value of 1 indicates that the autocorrelograms are identical, whereas larger values indicate less similarity. Ratios were around 1 if both autocorrelograms were from responses to wideband noise and were greater than 1 if one autocorrelogram was from responses to iterated rippled noises or harmonic complex tones. This ratio may be a useful decision variable for evaluating temporal coding schemes. [Work supported by NIDCD P01 DC00293.]