Fast coding in the primary visual cortex of the macaque monkey

Fast coding in the primary visual cortex of the macaque monkey Dr. Maoz Shamir, Ben-Gurion University of the Negev Dr. Adam Kohn Albert Einstein College of Medicine Final Report During the first phase of this research project we started to investigate the contribution of the temporal structure of the neural responses to the information content about the stimulus identity. Specifically we focused on the possible role of response latency in cases where fast and reliable decision about the external stimulus is essential.

In our preliminary theoretical investigation, we have suggested a means to quantify information content of response latency using a simple ʻrace-to-thresholdʼ readout mechanism. The accuracy of this readout has been analyzed theoretically based on abstract model for the neural response statics. However, the utility of this approach has not been put to the test on real neural data.

To this aim experiments were undertaken in the laboratory of Dr. Adam Kohn to harvest array data containing the simultaneous spiking activity of large nerve cell populations form the primary visual cortex of the monkey responding to visual gratings stimuli of variable orientations over hundreds of trials per stimulus condition.

The data from the laboratory of Dr. Adam Kohn was then analyzed in the theoretical laboratory of Dr. Maoz Shamir. We find that response latency is indeed tuned to the stimulus. Typically, response latency exhibits similar tuning structure as the conventional tuning of the rate, i.e. unimodeal tuning that peaks at the neuron's preferred orientation. Further analysis of electrophysiological data reveals that response latency can encode information about features of the external stimulus with sufficient accuracy to allow for accurate discrimination between a few alternatives. To extract information from neuronal response latency, estimation of stimulus onset is required. We show that cells that do not encode relevant stimulus features by their response latency provide a reliable and accurate estimate of stimulus onset time. These results have been presented in several scientific meetings and have been published in PLoS Computational Biology.

During the second stage of our research we investigated the effect of noise correlations and neuronal heterogeneity on the coding accuracy of several readout algorithms. We find that although the functional dependence of noise correlations can limit simple readout mechanisms such as the naïve population vector, a readout that takes into account the noise structure and neuronal heterogeneity can overcome the detrimental effect of correlated noise. These findings have been presented in scientific meetings and are currently summarized to an article.