A CoRD (Constructive Recursive Algoritm) for neural networks on the plane

CABRELLI, CARLOS; MOLTER, URSULA; SHONKWILER, RON

IEEE TRANSACTIONS ON NEURAL NETWORKS

IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC

Lugar: New York; Año: 2000 vol. 4 p. 345 - 356

1045-9227

A sufficient condition that a region be classifiable by a 2-layer feed-forward neural net (a two-layer Perceptron) using threshold activation functions is that either it be a convex polytope, or that intersected with the complement of a convex polytope in its interior, or that intersected with the complement of a convex polytope in its interior, or . . . recursively. These have been called Convex Recursive Deletion (CoRD) regions. We give a simple algorithm for finding the weights and thresholds, in both layers, for a feed forward net which implements such a region.