Temporal multidirectional associative memory: adaptable, continuous, and self-connected MAM

This paper introduces a new version of the multidirectional associative memory (MAM) in which the state of activation of the processing units ranges from -1 to +1 and each unit has self-connections. Furthermore, the correlation matrices, generated by Hebbian rules, are trained by Widrow-Hoff rule. The model aims at reproducing temporal sequences. The results of the experiments suggest that the model has a fast training stage, improves the learning capacity of MAM, reproduces trained temporal sequences, interpolates and extrapolates states in a trained sequence, and improves the accuracy of the recall with the inclusion of states