Next: INTRODUCTION
Probability Interpretation of Population Codes
Terence D. Sanger
Massachusetts Institute of Technology
Room E25-526
Cambridge, MA 02139
(617)253-5769 tds@ai.mit.edu
Preference: Oral
Topic: Theory; Distributed Representations
Abstract:
Although theoretical neural network models frequently assume that the
firing rates of individual neurons carry the analog signals used for
computations, it is becoming clear that in many sensory and motor systems,
information is encoded as patterns of activity over large populations of
cells [Georgopoulos et al. 1988, for example,]. I seek to address several different but
related questions
concerning such populations. I investigate the computational
properties of certain types of population codes by using a
probabilistic interpretation of their structure. I show the existence of
simple algorithms for nonlinear supervised learning as well as for
unsupervised learning based on information maximization
[Linsker 1988,Linsker 1989]. And I show that
although in general it is not possible to find a unique external variable
that is coded by a population [Sanger 1994], it is
possible to use information maximization principles
to find the external variable that is represented best by the
population.
Terence D. Sanger
Mon Aug 21 18:36:58 EDT 1995