48 min
Seminar, Hosted by Simons Institute for the Theory of Computing
Control of Synaptic Plasticity in Deep Cortical Networks
University of Amsterdam
Recording
Abstract
Pieter Roelfsema proposes that feedback from motor cortex identifies which synapses in association cortex should change when an action is rewarded. The architecture learns action values across tasks involving delayed and memory-guided eye movements, decisions and classification. The talk compares its predicted visual, memory and action-value responses in intermediate layers with neurophysiological evidence.