Presentation | 2009-01-19 Reinforcement meta-learning rule solves the distal reward problem Shojiro ARAKI, Yutaka SAKAI, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | It is known that spike-timing-dependent synaptic plasticity (STDP) depends on the initial strength of the synapse, and that the dependence is asymmetric for potentiation and depression. It is pointed out that this fact implies a problem: the destination of a synapse should be restricted in a small region detemined by the initial-strength dependence, and little reflect the input-output statistics. If it holds true, then the learning paradigm drawn by Hebb would be broken. In order to solve the problem, we proposed a meta-learning rule depending on reinforcement signals. We applied the meta-learning for STDP learning rule that possesses asymmetric initial-strength dependence, and demonstrated that a single model neuron can learn the selectivity reflecting input statistics. We assume that the reinforcement signals reflect rewards given to the animal, and spread over the whole brain. Here we demonstrate that a single model neuron can learn the selectivity reflecting inputs correlated with rewards given a few seconds after the inputs. The proposed reinforcement meta-learning can solve the distal reward problem as well as the problem in the initial-strength dependence. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | STDP / dopamine / meta-learning rule / distal reward |
Paper # | NC2008-95 |
Date of Issue |
Conference Information | |
Committee | NC |
---|---|
Conference Date | 2009/1/12(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Neurocomputing (NC) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Reinforcement meta-learning rule solves the distal reward problem |
Sub Title (in English) | |
Keyword(1) | STDP |
Keyword(2) | dopamine |
Keyword(3) | meta-learning rule |
Keyword(4) | distal reward |
1st Author's Name | Shojiro ARAKI |
1st Author's Affiliation | Graduate school of engineering, Tamagawa University() |
2nd Author's Name | Yutaka SAKAI |
2nd Author's Affiliation | Tamagawa University Brain Science Institute |
Date | 2009-01-19 |
Paper # | NC2008-95 |
Volume (vol) | vol.108 |
Number (no) | 383 |
Page | pp.pp.- |
#Pages | 5 |
Date of Issue |