OPTIMIZATION PRINCIPLES FOR THE NEURAL CODE

Recent experiments show that the neural codes at work in a wide range of creatures share some common features. At first sight, these observa tions seem unrelated. However, we show that these features arise natur ally in a linear filtered threshold crossing model when we set the thr eshold to maximize the transmitted information. This maximization proc ess requires neural adaptation to not only the DC signal level, as in conventional light and dark adaptation, but also to the statistical st ructure of the signal and noise distributions. We also present a new a pproach for calculating the mutual information between a neuron's outp ut spike train and any aspect of its input signal which does not requi re reconstruction of the input signal. This formulation is valid provi ded the correlations in the spike train are small, and we provide a pr ocedure for checking this assumption. This paper is based on joint wor k (DeWeese M 1995 Optimization principles for the neural code Disserta tion Princeton University). Preliminary results from the linear filter ed threshold crossing model appeared in a previous proceedings (DeWees e M and Bialek W 1995 Information flow in sensory neurons Nuovo Ciment o D 17 733-8), and the conclusions we reached at that time have been r eaffirmed by further analysis of the model.