Although individual sensory neocortical regions process very different types of information (visual, auditory, tactile), their anatomical construction and physiological operation share many prominent features, suggesting common behaviors that may underlie superficially different sensory processing functions. A model consisting of these shared features, using learning rules modeled after synaptic long-term potentiation (LTP), is shown to possess the capability to process continuous time-varying sequences via high-capacity storage and retrieval of temporal feature strings of arbitrary length. Operation of the model is illustrated via application to recognition of handwriting via left-to-right scanning of text images. The model constitutes a novel hypothesis of underlying functions of sensory neocortical circuitry which, it is argued, are similar across modalities. The model may also represent the beginnings of a novel approach to temporal signal processing.
展开▼
