1 h
Seminar, Hosted by Simons Institute for the Theory of Computing
Backpropagation and Deep Learning in the Brain
DeepMind Technologies Limited, London
Recording
Abstract
Timothy Lillicrap reviews how error backpropagation made it possible to train large systems for complex behavior, including in deep reinforcement learning. He then asks which cortical mechanisms could play a comparable role. Evidence for electrically segregated neuronal compartments and computation in dendrites motivates possible ways to deliver detailed learning signals in a biological network.