Himberger, K. & Honey, C.
University of Toronto, Department of Psychology
How does the brain integrate sequences of information that unfold over multiple timescales, from milliseconds to minutes? To understand the general basis of sequential information processing in local cortical circuits, we have begun by modeling sequence learning in the constrained setting of early visual cortex. Recent experiments have shown that, within primary visual cortex, neuronal responses to a sequence of four visual orientation stimuli (where each stimulus is presented for 150ms) are increased after repeated sequence exposures; the effect is specific to the sequence order and absolute sequence timing, and ?predictive? responses are observed when a sequence element is omitted. To model this phenomenon, we demonstrate how spike timing dependent plasticity (STDP), operating at individual synapses on short time scales (~20-50 ms), can nonetheless shape larger ensembles of neurons (synfire chains) to implicitly represent temporal statistics over hundreds of milliseconds. Finally, we consider how the mechanisms of spatiotemporal sequence learning in a sensory cortical circuit can generalize to spatiotemporal statistical learning in higher order regions of the cerebral cortex.