Co-adaptive BMI by reinforcement learning based on prediction of users' latent mental states

2014 Fiscal Year Final Research Report

• Project/Area Number: 24300093
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Partial Multi-year Fund
• Section: 一般
• Research Field: Sensitivity informatics/Soft computing
• Research Institution: Advanced Telecommunications Research Institute International
• Principal Investigator: KAWANABE Motoaki 株式会社国際電気通信基礎技術研究所, 認知機構研究所, 主任研究員 (30272389)
• Co-Investigator(Kenkyū-buntansha): KANEMURA Atsunori 独立行政法人産業技術総合研究所, ヒューマンライフテクノロジー研究部門, 研究員 (50580297) ; UENO Tsuyoshi 大阪大学, 産業科学研究所, 招聘研究員 (90615824)
• Co-Investigator(Renkei-kenkyūsha): WASHIO Takashi 大阪大学, 産業科学研究所, 教授 (00192815)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: ブレイン・マシン・インタフェース / 強化学習

Outline of Final Research Achievements

Toward a real-time co-adaptive BMI algorithm for providing flexible feedback schemes based on users' latent mental states, we developed reinforcement learning procedures to construct BMI decoders and representations for brain activities to infer the mental states. For the former topic, based on the weighted likelihood, we establish a theoretical framework to determine the optimal state modeling, namely the dimension of the mental states and their transition rule, to design an appropriate policy model, and to execute reinforcement learning simultaneously. For the latter topic, we proposed various generalizations of the standard feature extraction method CSP (common spatial pattern) to construct robust features against subject-to-subject variability and non-stationarity in brain signals. By integrating these element technologies, we implemented a BMI feedback device with portable EEG and a ball lamp, and tested its usefulness with a few subjects in a real-world environment.

Free Research Field

医用工学・機械学習