2A2-D11 強化学習による複雑ネットワーク上の行動獲得
スポンサーリンク
概要
- 論文の詳細を見る
We investigated a navigation learning problem in a certain class of complex networks called scale-free networks. A scale-free network is characterized as the distribution of degrees shows scale-free property. For some nodes that have large degree called "hub", exploration is more important. For nodes with less number of edges, however, action selection exploiting already obtained knowledge on the environment may be possible. The temperature parameter in the equation of the softmax selection can be used to regulate the degree of exploration and exploitation. In the present study, the learning agent learns to navigate in networks generated using the Barabasi-Albert model that have the scale-free property. Learning agents with various tendency of exploration and exploitation acquired distinctive navigation behaviors.
- 一般社団法人日本機械学会の論文
- 2009-05-25
著者
関連論文
- チャット会話の秩序:インターバル解析による会話構造の研究
- ヒラメ個体群の定向配列行動に出現する階層性(生命・進化・ゲーム,基研長期研究会「複雑系4」)
- 動物行動の可塑性をいかに理解するか
- 2A2-D11 強化学習による複雑ネットワーク上の行動獲得