COMPETITION BETWEEN PATTERN RECALL AND SEQUENCE PROCESSING IN A NEURAL-NETWORK STORING CORRELATED PATTERNS

We investigate the effects on a network that stores a sequence of patt erns without time delays of introducing patterns that are correlated i n a way that depends only on their separation from each other in the s equence. We demonstrate that with a carefully chosen form for the syna ptic matrix, the introduction of these correlations can, under certain circumstances, aid the recall of the individual patterns while still preserving the limit-cycle behaviour of the net.