An ART2-TPM neural network for automatic pattern classification
M. Fujita, Behnam Bavarian · 2002
Presents a novel two-layer neural network based on the adaptive resonance theory (ART2) network for continuous variables in which the bottom-up long-term memory (LTM) and the top-down LTM adaptations use the self-organizing topology preserving mapping (TPM) learning rule. This topology is developed in the context of an extended Neocognitron for automatic pattern clustering from raw two-dimensional image inputs to a finite number of classes. The complete ART2-TPM algorithm is presented and an illustrative example for automatic recognition of line orientation is given. The performance of the network is compared to that of the winner-take-all network.>