情報の取得を包含した制御理論と統計的学習理論の融合数理基盤

• Project/Area Number: 21H04875
• Research Category: Grant-in-Aid for Scientific Research (A)
• Allocation Type: Single-year Grants
• Section: 一般
• Review Section: Medium-sized Section 60:Information science, computer engineering, and related fields
• Research Institution: Kyoto University
• Principal Investigator: 加嶋 健司 京都大学, 情報学研究科, 准教授 (60401551)
• Co-Investigator(Kenkyū-buntansha): 蔡 凱 大阪公立大学, 大学院情報学研究科, 教授 (20728266) ; 河野 佑 広島大学, 先進理工系科学研究科(工), 准教授 (40743034)
• Project Period (FY): 2021-04-05 – 2026-03-31
• Project Status: Granted (Fiscal Year 2024)
• Budget Amount: ¥41,470,000 (Direct Cost: ¥31,900,000、Indirect Cost: ¥9,570,000); Fiscal Year 2025: ¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000); Fiscal Year 2024: ¥9,620,000 (Direct Cost: ¥7,400,000、Indirect Cost: ¥2,220,000); Fiscal Year 2023: ¥9,490,000 (Direct Cost: ¥7,300,000、Indirect Cost: ¥2,190,000); Fiscal Year 2022: ¥9,620,000 (Direct Cost: ¥7,400,000、Indirect Cost: ¥2,220,000); Fiscal Year 2021: ¥8,190,000 (Direct Cost: ¥6,300,000、Indirect Cost: ¥1,890,000)
• Keywords: 制御理論 / 統計的学習理論 / 機械学習 / 最適化

Outline of Research at the Start

潤沢なデータと機械学習を活用した制御システム設計は，非線形性や環境の変化への柔軟な対応力を有し，今後さらなる適用範囲の拡大が期待されている．一方で， IoT（Internet of Things）に代表される対象の大規模複雑化やプライバシーの問題により，今後は「いかにデータおよびドメイン知識を取得するか」が重要な課題となると予想される．そこで本研究では，情報の取得戦略も包括的に扱うことができる形でシステム制御理論と統計的学習理論を融合した基盤数理を創生し，取得前データの価値の定量化やデータ化が困難な「現場の知見」の系統的な活用のための方法論を確立することを目指す．

Outline of Annual Research Achievements

本研究では機械学習手法に簡単には組み込めないダイナミクス特有のドメイン知識を活用したモデリング手法を開発することを目指している。これに関して、本年度は、定常応答の事前情報を活用するダイナミクスモデリング、離散事象システムにおけるデータ駆動制御、動的システムにおけるエントロピーと最適輸送の関係性の解明と活用、分布情報にもとづくダイナミクスモデリング、などの理論構築を行った。例として、最適輸送に関する結果においては、静的な最適輸送問題に対して広く用いられているSinkhornアルゴリズムを、モデル予測制御という実用的な制御手法に応用し、さらにその安定性を証明することに成功した。この成果は、システム制御分野のトップジャーナルのひとつであるAutomaticaにおいてEditors' Choiceに選出されている。他にも、システム制御分野のトップジャーナルを多く含む９編の査読つき国際論文誌の出版や、計測自動制御学会論文賞を受賞するなど、国内外で広く成果が認められ始めている。また、これらの理論結果について、研究代表者が参画している未病発見に関するムーンショットプロジェクトやモビリティ社会に関する産学連携プロジェクトへの応用展開も平行して検討しており、成果を社会へ還元する一手法となることを期待している。また、Gromov-Wasserstein距離にもとづく最適輸送と確率制御の関連を明らかにする計画外の研究も実施し、一定の成果が得られた。この距離関数は異なるドメイン間の性質的差異を許容する特徴をもつため、本課題にとってもその意義は大きい。

Current Status of Research Progress

2: Research has progressed on the whole more than it was originally planned.

Reason

前項で述べたように、予定していた結果を順調に導出し、適切なタイミングで出版をおこなっている。

Strategy for Future Research Activity

本研究では機械学習手法に簡単には組み込めないダイナミクス特有のドメイン知識(以下では非データ型ドメイン知識と呼ぶ)を活用したモデリング手法を開発す る。本年度は、最適輸送理論、最大エントロピー制御に関する研究など多くの個別課題が着実に進んでいるため、継続的に論文投稿をおこなう。応用においては、昨年度まで雇用していたオピニオンダイナミクスを専門とする研究員との共同研究を継続し、ネットワークモデリング手法の有効性を検証する。 一方で、並行してCyber Physical Systemsなどの数理モデルにおいて頻出する微分/差分方程式では捉えづらい論理演算を内包するダイナミク ス(離散事象システ ム)に対するデータ駆動型システム制御理論の構築に関する研究も継続する。具体的には、前年度に世界に先駆けて国際会議で発表したデータ有用性の概念にもとづいて、 データ駆動設計手法を体系化・論文化をすすめる。さらに、前年度に計画外に実施し一定の成果が得られたGromov-Wasserstein距離にもとづく最適輸送と確率制御の関連を明らかにする研究に関して、実質的に最終年度となる本年度に発展可能性を検討し、次の研究課題申請へのステップとなるか検討する予定である。

Research Products

• [Int'l Joint Research] University of Groningen(オランダ)
• [Int'l Joint Research] University of Cagliari(イタリア)
• [Int'l Joint Research] Xidian University(中国)
• [Int'l Joint Research] Xidian University(中国)
• [Journal Article] Contraction analysis of discrete-time stochastic systems (2024)
  • Author(s): Yu Kawano, Yohei Hosoe
  • Journal Title: IEEE Transactions on Automatic Control Volume: 69 Issue: 2 Pages: 982-997
  • DOI: 10.1109/tac.2023.3283678
  • Peer Reviewed / Open Access
• [Journal Article] Maximum Entropy Optimal Density Control of Discrete-Time Linear Systems and Schrテカdinger Bridges (2024)
  • Author(s): Kaito Ito, Kenji Kashima
  • Journal Title: IEEE Transactions on Automatic Control Volume: 69 Issue: 3 Pages: 1536-1551
  • DOI: 10.1109/tac.2023.3305319
  • Peer Reviewed
• [Journal Article] Krasovskii and shifted passivity based approaches to mixed input/output consensus (2023)
  • Author(s): Yu Kawano, Michele Cucuzzella, Jacquelien M.A. Scherpen
  • Journal Title: IEEE Control Systems Letters Volume: 7 Pages: 1951-1956
  • DOI: 10.1109/lcsys.2023.3282887
  • Peer Reviewed / Open Access / Int'l Joint Research
• [Journal Article] Krasovskii and shifted passivity based output consensus (2023)
  • Author(s): Yu Kawano, Michele Cucuzzella, Shuai Feng, Jacquelien M.A. Scherpen
  • Journal Title: Automatica Volume: 155 Pages: 111167-111167
  • DOI: 10.1016/j.automatica.2023.111167
  • Peer Reviewed / Open Access / Int'l Joint Research
• [Journal Article] Combinatorial Optimization Approach to Client Scheduling for Federated Learning (2023)
  • Author(s): Tomohito Omori, Kenji Kashima
  • Journal Title: IEEE Control Systems Letters Volume: 7 Pages: 2203-2208
  • DOI: 10.1109/lcsys.2023.3285739
  • Peer Reviewed
• [Journal Article] Entropic model predictive optimal transport for underactuated linear systems (2023)
  • Author(s): Kaito Ito, Kenji Kashima
  • Journal Title: IEEE Control Systems Letters Volume: 7 Pages: 2761-2766
  • DOI: 10.1109/lcsys.2023.3288474
  • Peer Reviewed
• [Journal Article] Deep reinforcement learning for continuous-time self-triggered control with experimental evaluation (2023)
  • Author(s): Ran Wang, Kenji Kashima
  • Journal Title: Advanced Robotics Volume: 37 Issue: 16 Pages: 1012-1024
  • DOI: 10.1080/01691864.2023.2229886
  • Peer Reviewed
• [Journal Article] Transformational Supervisor Localization (2023)
  • Author(s): Thuijsman Sander、Cai Kai、Reniers Michel
  • Journal Title: IEEE Control Systems Letters Volume: 7 Pages: 1682-1687
  • DOI: 10.1109/lcsys.2023.3278248
  • Peer Reviewed / Int'l Joint Research
• [Journal Article] A TSP-Based Online Algorithm for Multi-Task Multi-Agent Pickup and Delivery (2023)
  • Author(s): Kudo Fumiya、Cai Kai
  • Journal Title: IEEE Robotics and Automation Letters Volume: 8 Issue: 9 Pages: 5910-5917
  • DOI: 10.1109/lra.2023.3301300
  • Peer Reviewed
• [Journal Article] An LMI framework for contraction-based nonlinear control design by derivatives of Gaussian process regression (2023)
  • Author(s): Kawano Yu、Kashima Kenji
  • Journal Title: Automatica Volume: 151 Pages: 110928-110928
  • DOI: 10.1016/j.automatica.2023.110928
  • Peer Reviewed
• [Journal Article] Observability Gramian for Bayesian Inference in Nonlinear Systems With Its Industrial Application (2023)
  • Author(s): Kunwoo Lee、Umezu Yusuke、Konno Kaiki、Kashima Kenji
  • Journal Title: IEEE Control Systems Letters Volume: 7 Pages: 871-876
  • DOI: 10.1109/lcsys.2022.3227452
  • Peer Reviewed / Open Access
• [Journal Article] Entropic model predictive optimal transport over dynamical systems (2023)
  • Author(s): Ito Kaito、Kashima Kenji
  • Journal Title: Automatica Volume: 152 Pages: 110980-110980
  • DOI: 10.1016/j.automatica.2023.110980
  • Peer Reviewed / Open Access
• [Journal Article] Sparse control node scheduling in networked systems based on approximate controllability metrics (2022)
  • Author(s): Ikeda Takuya、Kashima Kenji
  • Journal Title: IEEE Transactions on Control of Network Systems Volume: - Issue: 3 Pages: 1-1
  • DOI: 10.1109/tcns.2022.3154666
  • Peer Reviewed / Open Access
• [Journal Article] Optimal Secret Protections in Discrete-Event Systems (2022)
  • Author(s): Ma Ziyue、Cai Kai
  • Journal Title: IEEE Transactions on Automatic Control Volume: 67 Issue: 6 Pages: 2816-2828
  • DOI: 10.1109/tac.2021.3091438
  • Peer Reviewed
• [Journal Article] Consistent reduction in discrete-event systems (2022)
  • Author(s): Cai Kai、Giua Alessandro、Seatzu Carla
  • Journal Title: Automatica Volume: 142 Pages: 110333-110333
  • DOI: 10.1016/j.automatica.2022.110333
  • Peer Reviewed / Int'l Joint Research
• [Journal Article] Transformational Nonblocking Verification (2022)
  • Author(s): Thuijsman Sander、Reniers Michel、Cai Kai
  • Journal Title: IFAC-PapersOnLine Volume: 55 Issue: 28 Pages: 256-263
  • DOI: 10.1016/j.ifacol.2022.10.351
  • Peer Reviewed / Int'l Joint Research
• [Journal Article] Efficient Safe Control via Deep Reinforcement Learning and Supervisory Control ? Case Study on Multi-Robot Warehouse Automation (2022)
  • Author(s): Konishi Masahiro、Sasaki Tomotake、Cai Kai
  • Journal Title: IFAC-PapersOnLine Volume: 55 Issue: 28 Pages: 16-21
  • DOI: 10.1016/j.ifacol.2022.10.318
  • Peer Reviewed
• [Journal Article] Controller Reduction for Nonlinear Systems by Generalized Differential Balancing (2022)
  • Author(s): Kawano Yu
  • Journal Title: IEEE Transactions on Automatic Control Volume: 67 Issue: 11 Pages: 5856-5871
  • DOI: 10.1109/tac.2021.3124980
  • Peer Reviewed
• [Journal Article] Bayesian Differential Privacy for Linear Dynamical Systems (2022)
  • Author(s): Sugiura Genki、Ito Kaito、Kashima Kenji
  • Journal Title: IEEE Control Systems Letters Volume: 6 Pages: 896-901
  • DOI: 10.1109/lcsys.2021.3087096
  • NAID: 40022585991
  • Peer Reviewed / Open Access
• [Journal Article] On Resilient Supervisory Control Against Indefinite Actuator Attacks in Discrete-Event Systems (2022)
  • Author(s): Ma Ziyue、Cai Kai
  • Journal Title: IEEE Control Systems Letters Volume: 6 Pages: 2942-2947
  • DOI: 10.1109/lcsys.2022.3168926
  • Peer Reviewed / Int'l Joint Research
• [Journal Article] Kullback-Leibler control for discrete-time nonlinear systems on continuous spaces (2022)
  • Author(s): Kaito Ito, Kenji Kashima
  • Journal Title: SICE Journal of Control, Measurement, and System Integration Volume: Vol. 15 Issue: 2 Pages: 119-129
  • DOI: 10.1080/18824889.2022.2095827
  • Peer Reviewed / Open Access
• [Journal Article] Multiple sparsity constrained control node scheduling with application to rebalancing of mobility networks (2021)
  • Author(s): Ikeda Takuya、Sakurama Kazunori、Kashima Kenji
  • Journal Title: IEEE Transactions on Automatic Control Volume: - Issue: 8 Pages: 1-1
  • DOI: 10.1109/tac.2021.3115441
  • Peer Reviewed / Open Access
• [Journal Article] Deep Reinforcement Learning for Continuous-time Self-triggered Control (2021)
  • Author(s): Wang Ran、Takeuchi Ibuki、Kashima Kenji
  • Journal Title: IFAC-PapersOnLine Volume: 54 Issue: 14 Pages: 203-208
  • DOI: 10.1016/j.ifacol.2021.10.353
  • Peer Reviewed / Open Access
• [Journal Article] Structural controllability and time-to-control of directed scale-free networks with minimum number of driver nodes (2021)
  • Author(s): Imae Takanobu、Cai Kai
  • Journal Title: Systems & Control Letters Volume: 156 Pages: 105025-105025
  • DOI: 10.1016/j.sysconle.2021.105025
  • Peer Reviewed
• [Journal Article] Usability aware secret protection with minimum cost (2021)
  • Author(s): Matsui Shoma、Cai Kai
  • Journal Title: Nonlinear Analysis: Hybrid Systems Volume: 43 Pages: 101111-101111
  • DOI: 10.1016/j.nahs.2021.101111
  • Peer Reviewed
• [Presentation] Observability Gramian for Bayesian Inference in Nonlinear Systems With Its Industrial Application (2023)
  • Author(s): Kunwoo Lee, Yusuke Umezu, Kaiki Konno, Kenji Kashima
  • Organizer: The 2023 American Control Conference
  • Int'l Joint Research
• [Presentation] Sparse Network Identification for Consensus Discrete-Time Systems (2023)
  • Author(s): Takuya Ikeda, Kenji Kashima
  • Organizer: 22nd IFAC World Congress
  • Int'l Joint Research
• [Presentation] Resilience Evaluation of Entropy Regularized Logistic Networks with Probabilistic Cost (2023)
  • Author(s): Koshi Oishi, Yota Hashizume, Tomohiko Jimbo, Hirotaka Kaji, Kenji Kashima
  • Organizer: 22nd IFAC World Congress
  • Int'l Joint Research
• [Presentation] Reinforcement Learning for Discrete-Time Static Noisy State Feedback Control with Reward Estimation (2023)
  • Author(s): Ran Wang, Ye Tian, Kenji Kashima
  • Organizer: 22nd IFAC World Congress
  • Int'l Joint Research
• [Presentation] Maximum Entropy Density Control of Discrete-time Linear Systems with Quadratic State Cost (2023)
  • Author(s): Kaito Ito, Kenji Kashima
  • Organizer: 22nd IFAC World Congress
  • Int'l Joint Research
• [Presentation] Cost Robustness of Maximum Entropy Optimal Controller on Euclidean Input Space (2023)
  • Author(s): Yota Hashizume, Kenji Kashima
  • Organizer: 2023 International Symposium on Nonlinear Theory and its Applications (NOLTA2023)
  • Int'l Joint Research
• [Presentation] On the System Identifiability from Probability Distribution Data of Linear Systems Driven by Gaussian Noise (2023)
  • Author(s): Kohei Morimoto, Kenji Kashima
  • Organizer: 2023 International Symposium on Nonlinear Theory and its Applications (NOLTA2023)
  • Int'l Joint Research
• [Presentation] Data-Informativity for Data-Driven Supervisory Control of Discrete-Event Systems (2023)
  • Author(s): Tomofumi Ohtsuka, Kai Cai, Kenji Kashima
  • Organizer: 62nd IEEE Conference on Decision and Control (CDC2023)
  • Int'l Joint Research
• [Presentation] Combinatorial Optimization Approach to Client Scheduling for Federated Learning (2023)
  • Author(s): Tomohito Omori, Kenji Kashima
  • Organizer: 62nd IEEE Conference on Decision and Control (CDC2023)
  • Int'l Joint Research
• [Presentation] DIfferentiable Sparse Optimal Control (2023)
  • Author(s): Ryotaro Shima, Ryuta Moriyasu, Sho Kawaguchi, Kenji Kashima
  • Organizer: 62nd IEEE Conference on Decision and Control (CDC2023)
  • Int'l Joint Research
• [Presentation] Entropic model predictive optimal transport for underactuated linear systems (2023)
  • Author(s): Kaito Ito, Kenji Kashima
  • Organizer: 62nd IEEE Conference on Decision and Control (CDC2023)
  • Int'l Joint Research
• [Presentation] Online supervisory control for safe and adaptive human-in-the-loop warehouse logistic automation (2023)
  • Author(s): Makoto Kimura, Manato Fujimoto, Kai Cai
  • Organizer: 第66回自動制御連合講演会
• [Presentation] Smart nudge system using autonomous robots (2023)
  • Author(s): Makoto Kimura, Manato Fujimoto, Kai Cai
  • Organizer: 第66回自動制御連合講演会
• [Presentation] Optimal secret protection in discrete event systems with dynamic clearance levels (2023)
  • Author(s): Ziyue Ma, Kai Cai
  • Organizer: 22nd IFAC World Congress
  • Int'l Joint Research
• [Presentation] グラフニューラルネットワークを用いた非均質合意ネットワークのラベル推定 (2023)
  • Author(s): 後藤直紀, 加嶋健司
  • Organizer: 第66回自動制御連合講演会
• [Presentation] Sinkhorn MPC: Model predictive optimal transport over dynamical systems (2022)
  • Author(s): Ito Kaito、Kashima Kenji
  • Organizer: 2022 American Control Conference
  • Int'l Joint Research
• [Presentation] Secret Protections in Discrete-Event Systems with Minimum Costs (2022)
  • Author(s): Jiang Jiagang、Ma Ziyue、Cai Kai
  • Organizer: 2022 American Control Conference
  • Int'l Joint Research
• [Presentation] Optimal time-varying topology for network systems (2022)
  • Author(s): Ikeda Takuya、Kashima Kenji
  • Organizer: 13th Asian Control Conference
  • Int'l Joint Research
• [Presentation] Matrix Pontryagin principle approach to controllability metrics maximization under sparsity constraints (2022)
  • Author(s): Tomofumi Ohtsuka, Takuya Ikeda, Kenji Kashima
  • Organizer: 24th International Symposium on Mathematical Theory of Networks and Systems
  • Int'l Joint Research
• [Presentation] Maximum entropy optimal density control of discrete-time linear systems (2022)
  • Author(s): Kaito Ito, Kenji Kashima
  • Organizer: 24th International Symposium on Mathematical Theory of Networks and Systems
  • Int'l Joint Research
• [Presentation] Continuous-time MPC of an engine airpath system based on machine-learning assisted modeling (2022)
  • Author(s): Yusuke Sakaguchi, Ryuta Moriyasu, Masayuki Kusunoki, Kenji Kashima
  • Organizer: 10th IFAC Symposium on Robust Control Design
  • Int'l Joint Research