Recognition of Spatiotemporal Patterns by Nonmonotone Neural Networks.
Masahiko Morita, Satoshi Murakami · 1997
A neural network model that recognizes spatiotemporal patterns without expanding them into spatial patterns is presented. This model forms trajectory attractors in the state space of a fully recurrent network by a simple learning algorithm using nonmonotone dynamics. When a spatiotemporal pattern is inputted after learning, the network state is attracted to the corresponding learned trajectory and the incomplete part of the input pattern is restored in the input part of the model; at the same time, the output part indicates which spatiotemporal pattern is being inputted. In addition, this model can recognize the learned patterns correctly even if they are temporally extended or contracted. 1 Introduction Neural networks for pattern recognition usually deal with a spatiotemporal pattern by expanding it into a spatial pattern using time-delay filters [Tank and Hopfield, 1987] or multilayer delay units [Waibel, 1989]. This method, however, has some drawbacks such that the temporal lengt...