Latest News

About CAMQ

Information for Authors

Editorial Board

Browse CAMQ Online

Subscription and Pricing

CAMQ Contacts

CAMQ Home

 

Volume 10,     Number 1,     Spring 2002

 

INFORMATION DISTORTION AND NEURAL CODING
TOMÁS GEDEON, ALBERT E. PARKER AND ALEXANDER G. DIMITROV

Abstract. Our main interest is the question of how neural ensemble activity represents sensory stimuli. In this paper we discuss a new approach to characterizing neural coding schemes. It attempts to describe the specific stimulus parameters encoded in the neural ensemble activity and at the same time determines the nature of the neural symbols with which that information is encoded. This recently developed approach for the analysis of neural coding [7], [9] minimizes an intrinsic information theoretic cost function (the information distortion) to produce a simple approximation of a coding scheme, which can be refined as more data becomes available. We study this optimization problem. The admissible region is a direct product of simplices. We show that the optimal solution always occurs at a vertex of the admissible region. This allows us to reformulate the problem as a maximization problem on the set of vertices and develop an algorithm, which, under mild conditions, always finds a local extremum. We compare the performance of this new algorithm to standard optimization schemes on synthetic cases and on physiological recordings from the cricket cercal sensory system.

 

Download PDF Files
 
(Subscribers Only)

© 2005, Canadian Applied Mathematics Quarterly (CAMQ)