Models of the basal ganglia for sequencing and timing


May 18, 2016 - 1:00pm
Northwest 243
About the Speaker
Sean Escola (Columbia)

The generation of sequences of behaviors and the accurate estimation of time are both fundamental computations that the brain must solve. Indeed the latter may subserve the former by facilitating the precise selection of transition times between sequence elements. Increasingly, experimental evidence points to a role for the basal ganglia in sequencing and timing. For example, during sequential behaviors in rodents, striatal cells are observed to response phasically at the transition times between sequence elements, suggesting that these responses drive transitions. On the other hand, during interval timing tasks, striatal responses appear to uniformly tile time and to rescale with changes in the interval, suggesting a clock-like function. In this talk, I will use models of cortical, basal ganglia, and thalamic circuits to explore these phenomena, and to argue the following. 1) The basal ganglia interacting with the thalamocortical loop implements a “multiplicative recurrent neural network”, which has been shown in the machine learning community to have computational advantages over standard recurrent networks. 2) The thalamocortical projections play an integral role in sequence generation and this role should be experimentally observable by optogenetic manipulation of these projections while measuring noise correlations in cortex. 3) In the context of structured cortical input, striatal responses during interval timing tasks may arise via learning at the inhibitory synapses between medium spiny neurons in striatum. And 4) once learned, striatal interval timing can be reproduced and rescaled in the absence of structured input. Of note, this final result supports recent experimental evidence that cortex is not needed for the production of learned time intervals.