Cosyne 2008 Workshops
March 3-4, 2008
Snow Bird, Utah
Speaker Name
Florentin Woergoetter, Bernstein Center for Computational Neuroscience, Göttingen, Germany
Talk Title
Neural control and three-factor differential hebbian learning in behaving systems
Talk Abstract
It is a well known fact that Hebbian learning is inherently unstable because of its self-amplifying terms: the more a synapse grows the stronger the post-synapticactivity and therefore the faster the synaptic growth. This effect can be prevented by using a third factor which times the learning appropriately. It can be shown that such system exhibit rigorous convergence properties when embedded into a behavioral context. The biological counterpart of such a third factor is a neuromodulator which switches the learning on at a certain moment in time. We show in some behavioural robotics experiments that our three factor learning clearly outperforms classical Hebbian learning. Furthermore it is possible to analytically demonstrate that the most important classes of network learning - correlation based differential Hebbian learning and reward based temporal difference learning - are mathematically equivalent when timing the learning with such a third-factor modulatory signal.
