研究活動

雑誌論文(2010)

``Brain-like Computing Based on Distributed Representations and Neurodynamics," Ken Yamane and Masahiko Morita, New Generation Computing, 2010, accepted.[PDF]




サイバニクス拠点SRA成果報告会(2010).

poster04

Inference an action intention from surface EMG signals using neural networks, Ken Yamane (Special thanks: Hiroshi Kawata, Fumihide Tanaka and Masahiko Morita).




サイバニクス・プロジェクト研究・最終報告会(2010)

ニューラルネットを用いた表面筋電位からの腕の動作推定と自由度の高い入力インタフェースの開発, 山根健(アドバイザ:森田昌彦先生,田中文英先生,協力者:川田浩史さん).




博士論文(2009年度).

分散表現と神経力学系に基づく知的情報処理, 山根健.




サイバニクス拠点SRA成果報告会(2009).

非単調神経回路網を用いたパターンベースの状態予測モデル, 山根健.




システム情報工学研究科セミナー(2008).

複数ニューラルネットによる分散表現に基づく状態予測に関する研究,山根健.




2nd IWC(2008).

poster03

Brain-like processing by neurodynamical systems, Ken Yamane and Masahiko Morita, 2nd International Workshop on Cybernics.




サマースクールにおける研究紹介(2008).

poster02

Inference based on distributed representations using neural networks, Ken Yamane.




サイバニクス拠点RA成果報告会(2008).

poster01

Inference based on distributed representations using neurodynamics, Ken Yamane.




システム情報工学研究科セミナー(2007).

分散表現だけに基づいた推論システムの構築,山根健.




ICONIP2007(2007)

Inference based on distributed representations using trajectory attractors, Yamane, K., Hasuo, T. and Morita, M., Proceedings of the the 14th International Conference on Neural Information Processing.




修士論文(2006年度).

軌道アトラクタを用いたパターンベース推論,山根健.[簡単な内容]




雑誌論文(2007)

軌道アトラクタを用いたパターンベース推論,山根健・蓮尾高志・末光厚夫・森田昌彦,電子情報通信学会論文誌(D),vol.J90-D,no.3,pp.933-944,Mar. 2007.[PDF]




システム情報工学研究科・理工学研究科セミナー(2006).

軌道アトラクタモデルによるパターンベース推論,山根健.




第15回インテリジェント・システム・シンポジウム(2005).

非単調ニューラルネットによるパターンベース推論,山根健・末光厚夫・森田昌彦.
Abstract:The nonmonotone neural network, a fully recurrent network consisting of nonmonotonic neural elements, can form trajectory attractors in its state space, along which the network makes stable state transitions.Using this characteristic and a novel method of contextual modification, we can construct a pattern-based reasoning system that does not use any local representations or symbols.The present paper describes an example of such systems and shows that it can deal with exceptional data in a simple manner while maintaining strong generalization ability, and thus can make common-sense inferences from a limited amount of given rules.




システム情報工学研究科・理工学研究科セミナー(2005)

軌道アトラクタモデルによるパターンベース推論,山根健.




ニューロコンピューティング研究会(2005).

軌道アトラクタモデルによる分散表現に基づく推論,山根健・末光厚夫・森田昌彦.
Abstract:The trajectory attractor model is a recurrent neural network that can make continuous state transitions along learned trajectories, and is capable of simulating any large-scale finite automaton without using local representations or symbols.In the present paper, we show a system using this model that reasons based on distributed representation.This model not only has strong generalization ability but also can deal with exceptional data in a simple and natural manner, and thus can make common-sense inferences from a limited amount of given rules.




卒業研究(2004年度).

軌道アトラクタを用いた推論モデルの能力,山根健.
Abstract:The trajectory attractor model is a recurrent neural network that can make continuous state transitions along learned trajectories, and has drawn much attention recently since it can simulate any finite automaton without using local representations or symbols. In the present study, several kinds of simulation experiments were performed to examine the reasoning ability of this model. As a result, it was demonstrated that this model can deal with exceptional knowledge in a simple manner and makes a flexible inference;for example, it can infer plausible conclusions in a novel situation from a few learned rules. This indicates that the model fully uses the merits of distributed representation and is highly promising for constructing a human-like reasoning system.