Estimate of mutual information carried by neuronal responses from small data samples

Abstract

Calculation of information carried by neuronal responses evoked by external stimuli has proven to be an effective tool to study the behaviour of evoluted nervous systems. The major problem that arises in this analysis is that the number of stimulus/response pairs available from an experiment is limited, while the calculation of the true information would ideally require an infinite number of pairs. In this paper we present a method that allows an accurate estimate of the information from a limited number of pairs. The method is tested on two types of synthethic data. The first type is representative of a typical difficult problem in pattern recognition and signal detection, while the second type closely fits the response of geniculate parvocellular neurons to visual stimuli.