A Method for Temporal Association in Bayesian Networks

Abstract

A method for sequential pattern recognition and prediction in Bayesian networks is investigated. The basic approach in this method is to add stimulus delay lines to an associative network, thus converting temporal structure to a spatial one. Some methods to avoid very large connection matrices are studied. Results show that it is possible to efficiently store sequences in a network where the connection matrix is strongly reduced.


Authors:
Roland Orre, Anders Lansner
Last modified: Sun Feb 16 16:48:00 CET 2003