Measurement of Anticipative Power of a Retina by Predictive Information

Abstract

The predictive properties of a retina are studied by measuring the mutual information (MI) between its stimulation and the corresponding firing rates while it is being probed by a train of short pulses with stochastic intervals. Features of the measured MI at various time shifts between the stimulation and the response are used to characterize the predictive properties of the retina. By varying the statistical properties of the pulse train, our experiments show that a retina has the ability to predict future events of the stimulation if the information rate of the stimulation is low enough. Also, this predictive property of the retina occurs at a time scale similar to the well established anticipative phenomenon of omitted stimulus response in a retina. Furthermore, a retina can make use of its predictive ability to distinguish between time series created by an Ornstein-Uhlenbeck and a hidden Markovian process.