Speaker: Prof. S. Seung, Massachusetts Institute of Technology
Date: Wednesday, October 20, 2004
Time: 16:45 h (coffee/tea at 16:15 h in front of auditorium)
Place: ETH Hönggerberg, HPV G4
Title: Biologically plausible synaptic models for
reinforcement learning
Abstract:
Suppose that we find you practicing your serve on the tennis court, and
ask what is happening in your brain as you learn. If you are a
neuroscientist, you might reply that you are changing the synapses in
your brain to improve your performance. But if we were able to single
out a particular synapse in your brain, and ask whether strengthening it
would improve your serve, you would be unable to answer. Nevertheless,
knowledge of this sort must be available in your brain, if learning
indeed proceeds by changing the strengths of synapses. I will discuss
current theories of how the brain changes its synapses to optimize
behavior, and the prospects for testing these theories experimentally.