Neural Coding and Entropy 1 Conditional Stimulus and Spike Probabilities

C. Koch,EntropyChristof KochNovember

Abstract:The groundbreaking work of Kuuer (1953) at Harvard lead to a ood of studies concerning the receptive eld properties 1 of retinal ganglion cells, that is the region in visual space (in retinal coordinates) from which the cells' ring rate can be modulated. Rodieck (1965) was the rst to formally describe the receptive eld of cat ganglion cells using the diierence between two suitably scaled Gaussian functions, a narrow central Gaussian function and a much shallower and broader Gaussian. For an ON cell, the narrow Gaussian is excitatory and the broad one inhibitory and vice-versa for an OFF cell. Another closely related term from cellular physiology is the tuning curve, that is the response of the neuron as some stimulus parameter of relevance is varies, such as the contrast of the visual stimulus or the orientation of a line. The shape of 1-D tuning curves frequently falls into one of two categories, sigmoidal shaped, as the tuning for contrast or near and far-stereo cells for instance, and single-peaked Gaussian, as for direction of motion or orientation. Note that for the same cell, the tuning can be sigmoidal in one dimension and Gaussian is another. All of these methods characterize the response of the cell given some stimulus s(t). We can safely assume that to a good rst and second order approximation we can neglected anything about the individual spike except its time of occurence. Of course, we can't detect this time with arbitrary resolution but only within some reasonable level of resolution t (typically, t = 1 msec). This implies that we can characterize the entire response of the cell via a point process and only need to keep track of the times t i the impulses were generated. Typically, these times are written as series of delta function P i (t ? t i). In the following, we will abbreviate such a spike train as ft i g. Receptive elds and tuning curves then correspond to measuring the conditional probability P ft i gj s(t)] : (1) The most complete description of the behavior of the neuron as a function of the stimulus is of course the joint probability distribution P ft i g; s(t)] : (2) 1 A term introduced by Hartline at Rockefeller into the literature, working on the visual system of the horseshoe crab.