KATAMIC sequential associative memory

We present a neural architecture for storing, retrieving and recognizing multiple sets of sequences of patterns. This architecture provides an integrated mechanism for storage and recall. It is based on a novel neuron-like computing element, the prediction, which learns to generate at each time step a prediction of its next input based on its long-term experience. The model was implemented on a CM-2 connection machine and simulations were run to test the model for memory capacity, speed of training and accuracy of retrieval and recognition.