Presentation | 2015-03-16 Reinforcement Learning Based on Internal-Dynamics-Derived Exploration Using a Chaotic Neural Network Katsunari SHIBATA, Yuta SAKASHITA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | As a basic concept for emergence of intelligence through autonomous learning, exploration that is essential in reinforcement learning is considered as one aspect of the learner's internal dynamics, which is expected to develop through learning towards purposive and dynamic higher functions such as 'thinking'. A chaotic neural network is used to generate motions that include exploratory factors based on the chaotic dynamics without adding external random noises. Effective exploration is expected based on the dynamics such as "chaotic itinerancy", which is also the key to learn dynamic higher functions more easily that needs both stable and transitive dynamics. Furthermore, since exploration factors cannot be isolated in this internal-dynamic-derived exploration, a novel reinforcement learning method focusing on the correlation between the TD (Temporal Difference) error in the state value and the output change in each neuron is proposed. It was confirmed that an agent learned goal-directed behaviors in a simple task by reinforcement learning using a chaotic neural network. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | chaotic neural network / exploration / reinforcement learning / dynamics / thinking |
Paper # | MBE2014-166,NC2014-117 |
Date of Issue |
Conference Information | |
Committee | NC |
---|---|
Conference Date | 2015/3/9(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 Learning Based on Internal-Dynamics-Derived Exploration Using a Chaotic Neural Network |
Sub Title (in English) | |
Keyword(1) | chaotic neural network |
Keyword(2) | exploration |
Keyword(3) | reinforcement learning |
Keyword(4) | dynamics |
Keyword(5) | thinking |
1st Author's Name | Katsunari SHIBATA |
1st Author's Affiliation | Dept. of Electrical & Electronic Engineering, Oita University() |
2nd Author's Name | Yuta SAKASHITA |
2nd Author's Affiliation | Dept. of Electrical & Electronic Engineering, Oita University |
Date | 2015-03-16 |
Paper # | MBE2014-166,NC2014-117 |
Volume (vol) | vol.114 |
Number (no) | 515 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |