Spatio-temporal sequence processing with the counterpropagation neural network

We present a system that is capable of learning and retrieving spatio-temporal sequences using the forward-only counterpropagation neural network (FOCPNN). This artificial neural system has comparator units, a parallel array of FOCPNNs, and delayed feedback lines from the output of the system to the FOCPNN layer. The system has separate a conditioned stimulus (CS) input channel and unconditioned stimulus (US) input channel, which is analogous to classical conditioning. During learning, pairs of sequences of spatial patterns are presented to the CS and the US input channels simultaneously and the system learns to associate patterns at successive times in sequence. During retrieval, an imperfect cue sequence, which may be obscured by spatial noise and temporal gaps, causes the system to output the stored spatio-temporal sequence. Compared with other existing temporal systems, this system shows computational advantages such as fast and accurate learning and retrieving, and ability to store a large number of complex sequences consisting of non-orthogonal spatial patterns