A FEEDFORWARD NEURAL NETWORK CLASSIFIER MODEL: MULTIPLE CLASSES, CONFIDENCE OUTPUT VALUES, AND IMPLEMENTATION
Cris Koutsougeras, George M. Georgiou, C. Papachristou · International Journal of Pattern Recognition and Artificial Intelligence · 1992
The Athena model is a tree-like net for pattern classification. This paper presents the formalisms on which the model's internal representations and function are based. It also presents an adaptive algorithm to be used with this model. The adaptation is based on entropy optimization. The difficult problem of the optimization is handled by use of Fisher's multiple discriminants method. A method is also presented by which confidence values are produced for the overall classification decision. Finally, a data flow architecture using optical processing elements is considered for the model's implementation.