Reliability of neuronal responses to dynamic natural images in cat area 17

Jean-Philippe Lachaux, Shih-Cheng Yen, Jonathan Baker, Bruno A. Olshausen, Charles M. Gray.

Center For Neuroscience, UC Davis. Davis, California.

Abstract

Object recognition is thought to require some form of response invariance. Successive presentations of a stimulus should elicit, in some sub-population of neurons, a stable pattern of activation. It is not clear whether such patterns depend only on the average firing rate or whether the precise timing of spikes is important. This issue is unresolved for stationary stimuli, but dynamic stimuli must depend on some temporal modulation of the rate. In this context, it becomes important to know the temporal precision and reliability of sensory-evoked activity.

We estimated the reliability of neuronal activity over a range of time scales in response to movies. Single-unit recordings were performed in area 17 of anesthetized/paralyzed cats. Stimuli consisted of 60 repetitions of 30s natural movie sequences. For each movie and cell, we measured the firing rate as a function of time using bin sizes ranging from 4 to 100 ms. We then assessed the reliability of both the rate (using the Fano factor) and its time course. The time-course of the rate was evaluated on sliding windows of 20 consecutive bins; and the reliability of these waveforms was defined for each window as their correlation across the 60 repetitions of the stimulus.

We observed significant correlations of the rate time course for all time scales, including bin widths as small as 4ms. The latter correlations were episodic and typically occurred in response to rapid stimulus transients. These findings demonstrate that the temporal resolution of neuronal activity is not constant, but may adapt to the dynamic properties of the stimulus.

Supported by NEI.