Abstract

The interpattem-association (IPA) neuralnetwork model emphasizes the effect of the special features,¹ which makes it particularly suitable for application to reference patterns that are similar to one another. To achieve robustness of the neural network, some interconnection redundancy must be introduced. The IPA algorithm proposed in our previous publication is based on logic rules that determine the excitory and inhibitory interconnection. However, the interconnections were excessively redundant, and common features were not effectively suppressed.

© 1990 Optical Society of America