Measuring the Information Expressed by Neural Discharge Patterns

Abstract

Various measures have been used to assess how well single neuronsrepresent information. Modeling discharge patterns as stochastic pointprocesses, we determine how well certain measure accomplish this task.We show that the information theoretic measure-capacity-can do a poorjob. The mean-squared error measure more accurately describes thefidelity to which sensory signals can be extracted. Calculation offundamental bounds on mean-squared error show that time-varying signalsmust have bandwidths orders of magnitude less than the average dischargerate (under a Poisson model) if accurate signal representations are toresult. This result indicates that neural ensembles must be consideredto understand information encoding by neurons.