Sequence learning and timing in hippocampus, prefrontal cortex, and accumbens

Abstract

A basic architecture inspired from dentate gyrus and CA3-CA1 hippocampal fields combines a spectral timing module and an association network learning event transitions. According to the type of input the system can learn and replay: purely temporal sequences of aperiodic events; place-field chains as building blocks of graphs and maps, by combining visual and path-integration inputs; imitated sequences of movements by combining optic flow and movement-related proprioceptive feedback. The model is part of a triptych featuring also place cell computation and planning. The integrated architecture is used as a control system for robot navigation, sequence learning, prediction and novelty detection.