An artificial neural network system for temporal-spatial sequence processing

摘要

An artificial network is described that can learn, recognize, and generate higher-order temporal-spatial sequences. It consists of three parts: (1) comparator units, (2) a parallel array of artificial neural networks that are derived from the visual-vestibular networks of the snail Hermissenda, as well as hippocampal neuroanatomy, and (3) delayed feedback lines from the output of the system to the neural network layer. Its advantages include short training time, fast and accurate retrievals, toleration of spatial noise and temporal gaps in test sequences, and ability to store a large number of temporal sequences consisting of non-orthogonal spatial patterns.