Learning Reward Timing in cortex using reinforced expression of synaptic plasticity

From Santa Fe Institute Event Wiki

By Harel Shouval, TMC.

The ability to represent time is an essential component of cognition but its neural basis remains unknown. It is commonly believed that the underlying cellular mechanisms reside in high order cortical regions but recent studies show sustained neural activity in primary sensory cortices that are reward dependent and can represent the learned time of expected reward. Although extensively studied in both behavioral and electrophysiological studies, a general theoretical framework capable of describing the elementary neural mechanisms used by the brain to adaptively learn temporal representations is lacking. We present a model by which the brain can learn temporal representations through a simple theoretical framework predicated on reward dependent expression of synaptic plasticity. The model asserts that temporal representations are stored within the lateral neuronal connection matrix. We show that in the network model the representation can be accomplished by setting the connectivity matrix to have appropriate eigenvectors and eigenvalues. We further demonstrate that a simple model of reward dependent expression of neuronal plasticity is sufficient to learn the appropriate temporal representations. The model is successfully implemented in a network of stochastic spiking and we suggest experimentally verifiable predictions to test it. We also show how a population of such neurons can estimate the reward time given the spiking activity of the network, and that the error of this estimate scales linearly with the reward time, as shown experimentally.

Back to Agenda.