Competition between pattern reconstruction and sequence processing in nonsymmetric neural networks
A C C Coolen, David C. Sherrington · Journal of Physics A Mathematical and General · 1992
The authors study an Ising spin neural network model in which the interaction matrix consists of a symmetric Hebbian term (which favours the reconstruction of static patterns) and a nonsymmetric transition term (which favours limit cycles corresponding to the processing of pattern sequences). They calculate phase diagrams and analyse the relation between the relative weight of the two competing contributions to the interaction matrix and the frequency of the periodic attractors.