Probabilistic Encoding and Decoding of Neural Activity in Rat Primary Visual Cortex October 2002

Abstract

According to the information processing paradigm in the Cogni-tive Sciences, one of the nervous system's most important functions is to encode information about the environment. Understanding the neural code means understanding the relationship between brain states and real events in the outer world. In this work, a probabilistic framework for encoding (i.e., mapping events to neural response) and decoding (i.e., reconstruction of events from neural responses) of neural activity is presented and applied to theoretical and empirical data. Reconstruction is based on a Bayesian filter method, allowing the propagation of multi modal densities and the use of non-linear stimulus dynamics. The model is tested against empirical data recorded from rat primary visual cortex. Cells in this area are reported to be orientation-selective and their neural response characteristics can be learned by presenting oriented gratings to the rat's eye. The learned firing model is then used to reconstruct an unknown random walk stimulus from neural activity.