The resonance correlation network

The resonance correlation network (RCN) is a parallel continuous-time unsupervised pattern categorization machine than can operate on binary or continuous-valued data. The architecture is motivated by adaptive resonance theory (ART). Simple modifications, however, eliminate the problems associated with learning patterns presented for arbitrary time durations. The architecture also eliminates the sequential searching through previously coded patterns that is characteristic of ART-1. The RCN also increases its discrimination sensitivity as the learning process proceeds. The architecture and categorization properties of the resonance correlation network are described.<>