Abstract
A neural network model in which the connection matrix is formed by summing direct products of successive patterns in cyclic sequences is studied. It is found that the noise elimination performance of this model can be better than that of an autocorrelation model due to a reduced tendency for trapping in spurious attractors.