Predictive learning of temporal sequences in recurrent neocortical circuits.

Abstract

When a spike is initiated near the soma of a cortical pyramidal neuron, it may back-propagate up dendrites toward distal synapses, where strong depolarization can trigger spike-timing dependent Hebbian plasticity at recently activated synapses. We show that (a) these mechanisms can implement a temporal-difference algorithm for sequence learning, and (b) a population of recurrently connected neurons with this form of synaptic plasticity can learn to predict spatiotemporal input patterns. Using biophysical simulations, we demonstrate that a network of cortical neurons can develop direction selectivity similar to that observed in complex cells in alert monkey visual cortex as a consequence of learning to predict moving stimuli.