STIMULUS CODING THROUGH INTERSPIKE INTERVAL STATISTICS OF NEURAL POPULATIONS

Peter Cariani

Eaton Peabody Laboratory of Auditory Physiology

Massachusetts Eye and Ear Infirmary

Boston, MA

We will present an overview of the problem of neural coding of sensory information and present the example of the interspike interval coding of pitch and timbre in the auditory system. The first part of the talk will address general aspects of the neural coding problem as it is applied to problems of perception. There are several approaches. Systems identification techniques characterize response properties of neural elements (response prediction), information-theoretic approaches determine stimulus-related information content in spike trains (stimulus retrodiction), while perceptual coding approaches determine which aspects of neural activity covary with perceptual capabilities (psychoneural correspondence). Many possible neural codes can be contemplated. A coding space arises from the ways which information can be conveyed through neural ensemble activity: 1) which channels are activated (labelled lines, "place" codes), 2) how much they are activated (rate codes), 3) temporal patterns of spikes (interspike interval codes), and 4) time-of-arrival codes (relative latencies, interneural synchronies). Potential codes also include temporal and spatial sequences and joint occurences of events. Combinations of codes are common (rate-channel, latency-place). Stimulus coding can be achieved through extrinsically-impressed response patterns (mass statistics of stimulus-driven structure) or through activation of intrinsic response patterns (e.g.stimulus-specific impulse response shapes).

The coding of pitch in the auditory system provides a salient example of how stimulus qualities can be encoded through extrinsic, stimulus-driven temporal correlations between spikes. In the auditory nerve and cochlear nucleus, acoustic stimuli impress their time structure on the responses of many neurons, such that all-order interspike interval distributions reflect the correlation structure of the stimulus as it presents itself after cochlear filtering. Observed interspike interval distributions

from 50-100 fibers with different characteristic frequencies are summed together to form an estimate of the population-interval distribution of the auditory nerve. Population-interval distributions for many different stimuli that produce the same pitch (metameric stimuli allow one to hold the percept constant while varying the stimulus) are compared and different aspects of these distributions are evaluated with respect to how well they predict pitch judgements (and various "illusions"). Almost without exception, the most common interval in the population-interval distribution predicts the pitch that is heard (to within 1%), and the relative fraction of pitch-related intervals amongst all others qualitatively predicts the salience (strength) of that pitch. What is striking about these population-interval representations is that they are purely temporal they are based on temporal correlations between spikes rather than on which neurons have produced how many spikes. In population-interval distributions, all of the information about the identities of particular fibers and their tunings has been thrown away, yet the resulting sensory representation is highly accurate and extremely robust. Secondly, this is a clear example of a population-based neural code in which perceptual qualities are determined by asynchronous, temporal micropatterns of activity distributed over entire ensembles (mass statistics, correlations), rather than local activations of particular subsets of neurons (switchboards, across neuron patterns). Stimulus coding by temporal correlation is potentially available in any sensory system that phase-locks to its adequate stimulus (auditory, mechanoception, vision, electroception). Time and interest permitting, we will also discuss neural architectures (coincidence arrays, neural timing nets) that analyze these kinds of sensory representations.

References:

Cariani, P. 1999. Temporal coding of periodicity pitch in the auditory

system: an overview. Neural Plasticity 6(4):147-172.

Cariani, Peter A., and Bertrand Delgutte. 1996a. Neural correlates of

the pitch of complex tones. I. Pitch and pitch salience. J.

Neurophysiol. 76(3):1698-1716.

Cariani, Peter A., and Bertrand Delgutte. 1996b. Neural correlates of

the pitch of complex tones. II. Pitch shift, pitch ambiguity,

phase-invariance, pitch circularity, and the dominance region for pitch.

J. Neurophysiol. 76(3):1717-1734.