An algorithm derived from thalamocortical circuitry stores and retrieves temporal sequences

Different sensory neocortical architectures share prominent architectural and operational features, including convergent feedfoward and feedback connections with both specific and nonspecific thalamic nuclei, and synaptic long-term potentiation (LTP), a suspected substrate of learning. We present a subcircuit that is composed of a broad set of these shared constituents, and that is thus common to multiple neocortical regions. The subcircuit is shown to possess the capability for high-capacity storage and recognition of arbitrary-length temporal feature sequences. Operation of the circuit is illustrated here via its application to handwritten text recognition. The model constitutes a novel hypothesis of underlying functions of sensory neocortical circuitry which, it is argued, are similar across modalities. It is worth noting that the model also suggests a novel approach to handwriting recognition, requiring no preprocessing, no size normalization, and no segmentation, as well as having low space and time complexity costs.