Derivation of Natural Stimulus Feature Set Using a Data-Driven Model

Alexander G. Dimitrov, Tomas Gedeon, Brendan Mumey, Ross Snider, Ross Snider, Albert E. Parker, John P. Miller

Center for Computational Biology, Montana State University, Bozeman MT 59717, U.S.A.
Dimitrov@cns.montana.edu
Gedeon@cns.montana.edu
Mumey@cns.montana.edu
Snider@cns.montana.edu
Aldworth@cns.montana.edu
Parker@cns.montana.edu
Miller@cns.montana.edu

Abstract. A formal approach for deciphering the information contained within nerve cell ensemble activity patterns is presented. Approximations of each nerve cell's coding scheme is derived by quantizing its neural responses into a small reproduction set, and minimizing an information-based distortion function. During an experiment, the sensory stimulus world presented to the animal is modified to contain a richer set of relevant features, as those features are discovered. A dictionary of equivalence classes is derived, in which classes of stimulus features correspond to classes of spike-pattern code words. We have tested the approach on a simple insect sensory system.

LNCS 2660, pp. 337-345.

Last modified: