INFORMATION THEORETIC ANALYSIS OF NEURAL CODING AND PROCESSING

Don H. Johnson

Computer and Information Technology Institute

Departments of Electrical & Computer Engineering and Statistics

Rice University

Most current spike train analysis methods, be they single or multi-unit recordings, rely on point process models of the discharge patterns.  Such techniques focus on analyzing spike timing, but do not reveal much about how well the discharge patterns represent information and how well neural systems process information (perform feature extraction, for example).  I describe techniques derived from a new theory of information processing that quantify these questions.  We analyze recordings from central auditory pathways to show that neural coding can be dynamic and multi-faceted and from peripheral visual systems that neural information processing lossy can be (and must be so because of physiological constraints). We also show that population codes are capable of perfectly representing information.