Ring unidirectional associative memories

A novel architecture for associative memories, called ring unidirectional associative memories (RUAM), which can be used to store a number of pattern sequences, is presented. The RUAM could be used to recover learned pattern sequences from their noisy versions. The RUAM can be trained by both the Hebbian rule and by the backpropagation algorithm. Simulation results on noisy pattern sequence recall capability, storage capacity, and parameters' effects on convergence speed are presented. The backpropagation algorithm has stronger noisy pattern sequence recall capability and larger storage capacity than the Hebbian rule when it is used to train a RUAM to store nonorthogonal pattern sequences.<<ETX>>