ブラックボックス最適化研究室Black-Box Optimization Lab · 研究成果Research Findingsブラックボックス最適化研究室Black-Box Optimization Lab · 研究成果Research Findings

応用事例と共同研究Applications and Collaborations

国内外の大学や企業と共同で,船舶の自動運航,クローキングデバイスや構造のトポロジー最適化,CO₂ 地中貯留,データ分析など,さまざまな応用に私達の最適化技術を用いています.Together with universities and companies in Japan and abroad, we apply our optimization techniques to a range of applications, including autonomous ship navigation, topology optimization of cloaking devices and structures, CO₂ geological storage, and data analysis.

船舶の自動着桟・離桟と操縦運動モデルの同定(大阪大学 牧研究室)Autonomous Berthing and Unberthing of Ships and Identification of Maneuvering Models (Maki Lab, The University of Osaka)

キーワード:進化計算,最適制御,システム同定,ロバスト最適化Keywords: Evolutionary Computation, Optimal Control, System Identification, Robust Optimization

大阪大学の牧研究室と,船舶の自動運航に向けた共同研究を続けています.港内での着桟・離桟は,低速で風や潮流の影響を受けやすく,周囲の岸壁や他の船との衝突を避けながら操船する必要がある,自動化の難しい場面です.

[JMST 2020, JMST 2021b] では,着桟の最適制御問題を CMA-ES で解く方法を提案し,[JMST 2021a] では同じ考え方を追い波中で船が針路を保てなくなり転覆につながるおそれのあるブローチングの防止に適用しました.[Ocean Eng. 2022] では,実際の港の形状を考慮した着桟・離桟の軌道と制御の計画を最適化しています.制御の設計には,船の動きを表す操縦運動モデルが必要です.[JMST 2022, JMST 2024] では,操縦運動モデルのパラメータを CMA-ES で同定する方法や,港内操船のモデル化を自動化する方法を開発しました.さらに,風や操縦運動モデルの誤差があっても衝突を避けられる制御則を,最悪ケースの性能を最適化する方法で求めています [ACM TELO 2022](詳しくは最悪ケース・リスク・二段階の最適化).

We have a long-standing collaboration with the Maki Lab at The University of Osaka toward autonomous ship navigation. Berthing and unberthing in a harbor are difficult to automate: ships move at low speed, are strongly affected by wind and currents, and must maneuver while avoiding collisions with quays and other ships.

In [JMST 2020, JMST 2021b], we proposed solving the optimal control problem of berthing with CMA-ES, and in [JMST 2021a], we applied the same idea to preventing broaching-to, in which a ship in following seas loses its heading and may capsize. In [Ocean Eng. 2022], we optimized the trajectory and control plans for berthing and unberthing in a realistic port geometry. Control design requires a maneuvering model describing the ship's motion; in [JMST 2022, JMST 2024], we developed methods that identify the parameters of maneuvering models with CMA-ES and that automate the modeling of harbor maneuvers. Furthermore, we obtain control laws that avoid collisions despite wind and errors in the maneuvering model by optimizing the worst-case performance [ACM TELO 2022] (see Worst-Case, Risk-Aware, and Bilevel Optimization).

関連する論文Related Publications

  1. JMST 2020 JCR Q2 Scopus Q2
    Application of optimal control theory based on the evolution strategy (CMA-ES) to automatic berthing
    Atsuo Maki, Naoki Sakamoto, Youhei Akimoto, Hiroyuki Nishikawa, Naoya Umeda
    Journal of Marine Science and Technology, Vol.Vol. 25, pp. 221–233 (2020)
  2. JMST 2021a JCR Q2 Scopus Q2
    On broaching-to prevention using optimal control theory with evolution strategy (CMA-ES)
    Atsuo Maki, Naoki Sakamoto, Youhei Akimoto, Yuki Banno, Sreenath Maniyappan, Naoya Umeda
    Journal of Marine Science and Technology, Vol.Vol. 26, pp. 71–87 (2021)
  3. JMST 2021b JCR Q2 Scopus Q2
    Application of optimal control theory based on the evolution strategy (CMA-ES) to automatic berthing (Part: 2)
    Atsuo Maki, Youhei Akimoto, Naoya Umeda
    Journal of Marine Science and Technology, Vol.Vol. 26, pp. 835–845 (2021)
  4. Ocean Eng. 2022 JCR Q1 Scopus Q1
    Optimization on Planning of Trajectory and Control of Autonomous Berthing and Unberthing for the Realistic Port Geometry
    Yoshiki Miyauchi, Ryohei Sawada, Youhei Akimoto, Naoya Umeda, Atsuo Maki
    Ocean Engineering, Vol.Vol. 245, Art. 110390 (2022)
  5. JMST 2022 JCR Q2 Scopus Q2
    System Parameter Exploration of Ship Maneuvering Model for Automatic Docking / Berthing using CMA-ES
    Yoshiki Miyauchi, Atsuo Maki, Naoya Umeda, Dimas M. Rachman, Youhei Akimoto
    Journal of Marine Science and Technology, Vol.Vol. 27, pp. 1065–1083 (2022)
  6. JMST 2024 JCR Q2 Scopus Q1
    Development of Mathematical model for Harbor-Maneuvers to Realize Modeling Automation
    Yoshiki Miyauchi, Youhei Akimoto, Atsuo Maki
    Journal of Marine Science and Technology, Vol.Vol. 29, pp. 975–999 (2024)
  7. ACM TELO 2022
    Saddle Point Optimization with Approximate MinimizationOracle and its Application to Robust Berthing Control
    Youhei Akimoto, Yoshiki Miyauchi, Atsuo Maki
    ACM Transactions on Evolutionary Learning and Optimization, Vol.Vol. 2(1), Art. 2, pp. 1–32 (2022)

クローキングデバイスのトポロジー最適化(信州大学 藤井研究室)Topology Optimization of Cloaking Devices (Fujii Lab, Shinshu University)

キーワード:進化計算,トポロジー最適化,メタマテリアルKeywords: Evolutionary Computation, Topology Optimization, Metamaterials

信州大学の藤井研究室と,熱・電流・音・光などの流れを物体のまわりで迂回させ,物体をあたかも存在しないように見せるクローキングデバイスの設計に取り組んでいます.デバイスの形状(どこに材料を置くか)を最適化するトポロジー最適化は,一般には勾配法で解かれますが,目的関数が複雑な場合や設計の表現によっては勾配を使うことが難しくなります.

[CMAME 2018] では,レベルセット法による境界の表現と CMA-ES を組み合わせた構造のトポロジー最適化の方法を提案し,光学クロークとカーペットクロークに適用しました.この方法を用いて,熱 [APL 2018],直流電流 [Opt. Lett. 2019],音響と弾性の連成系 [APL 2021],電磁波と音の両方 [Opt. Express 2022] など,さまざまな物理現象に対するクローキングデバイスを設計しています.

With the Fujii Lab at Shinshu University, we design cloaking devices that guide flows of heat, electric current, sound, or light around an object so that the object appears not to exist. Topology optimization, which optimizes the shape of a device (where to place material), is usually solved by gradient methods, but gradients can be hard to use for complex objectives or some design representations.

In [CMAME 2018], we proposed a structural topology optimization method that combines a level-set boundary representation with CMA-ES, and applied it to optical and carpet cloaks. Using this method, we have designed cloaking devices for various physical phenomena, including heat [APL 2018], direct current [Opt. Lett. 2019], coupled acoustic-elastic systems [APL 2021], and both electromagnetic waves and sound [Opt. Express 2022].

関連する論文Related Publications

  1. APL 2018 JCR Q1 Scopus Q1
    Exploring optimal topology of thermal cloaks by CMA-ES
    Garuda Fujii, Youhei Akimoto, Masayuki Takahashi
    Applied Physics Letters, Vol.Vol. 112(6), Art. 61108 (2018)
  2. CMAME 2018 JCR Q1 Scopus Q1
    CMA-ES-based structural topology optimization using a level set boundary expression—Application to optical and carpet cloaks
    Garuda Fujii, Masayuki Takahashi, Youhei Akimoto
    Computer Methods in Applied Mechanics and Engineering, Vol.Vol. 332, pp. 624–643 (2018)
  3. Opt. Lett. 2019 JCR Q1 Scopus Q1
    DC carpet cloak designed by topology optimization based on CMA-ES
    Garuda Fujii, Youhei Akimoto
    Optics Letters, Vol.Vol. 44(8), pp. 2057–2060 (2019)
  4. APL 2021 JCR Q1 Scopus Q1
    Acoustic cloak designed by topology optimization for acoustic–elastic-coupled systems
    Garuda Fujii, Masayuki Takahashi, Youhei Akimoto
    Applied Physics Letters, Vol.Vol. 118, Art. 101102 (2021)
  5. Opt. Express 2022 JCR Q1 Scopus Q1
    Electromagnetic--acoustic biphysical cloak designed through topology optimization
    Garuda Fujii, Youhei Akimoto
    Optics Express, Vol.Vol. 30(4), pp. 6090–6106 (2022)

構造のトポロジー最適化(本田技術研究所)Structural Topology Optimization (Honda R&D)

キーワード:進化計算,トポロジー最適化,マルチフィデリティ最適化Keywords: Evolutionary Computation, Topology Optimization, Multi-Fidelity Optimization

本田技術研究所と,構造部材のトポロジー最適化に取り組んでいます.高精度なシミュレーションは計算時間が長く,試せる設計の数が限られることが課題です.[PPSN 2020] では,精度の異なるシミュレーションを順位相関に基づいて自動で切り替える方法(これまでの研究を参照)を,設計の事前・事後の制約をもつコンプライアンス最小化問題に適用しています.

With Honda R&D, we work on topology optimization of structural components. High-fidelity simulations are time-consuming, which limits the number of designs that can be evaluated. In [PPSN 2020], we applied our method that automatically switches between simulations of different fidelity based on rank correlation (see Past Research) to compliance minimization with prior and posterior constraints on the design.

関連する論文Related Publications

  1. PPSN 2020 CORE A
    Multi-Fidelity Optimization Approach under Prior and Posterior Constraints and its Application to Compliance Minimization
    Youhei Akimoto, Naoki Sakamoto, Makoto Ohtani
    Parallel Problem Solving from Nature – PPSN XVI. PPSN 2020. Lecture Notes in Computer Science, vol 12269, pp. 81–94 (2020)

CO₂ 地中貯留の坑井配置の最適化(大成建設)Well Placement Optimization for CO₂ Geological Storage (Taisei Corporation)

キーワード:進化計算,ロバスト最適化,混合整数最適化Keywords: Evolutionary Computation, Robust Optimization, Mixed-Integer Optimization

大成建設と,CO₂ を回収して地中に貯留する CCS(Carbon dioxide Capture and Storage)のための坑井の配置の最適化に取り組んでいます.地層のシミュレーションは計算コストが高く,地下の性質には不確かさがあります.[GECCO 2018] では,坑井の数と位置を同時に決める混合整数の問題として定式化して CMA-ES で解き,[GECCO 2019] では,地質統計的に生成した複数の地層のシナリオに対して性能を保証する配置を求めました.[ASC 2023] では,最悪のシナリオでの性能を効率よく推定しながら探索する方法を提案して,坑井配置の最適化に適用しています.また,並列計算による高速化を含む最適化ツールを開発しています [JSCE 2021].共同研究の成果である,CO₂ を効率よく地中に圧入するための井戸配置の最適化技術は,2026 年 10 月に大成建設から発表され,日経電子版,建設工業新聞,建設通信新聞で報道されました(大成建設のニュースリリース,日経電子版).

With Taisei Corporation, we optimize the placement of wells for carbon dioxide capture and storage (CCS). Reservoir simulations are computationally expensive, and the subsurface properties are uncertain. In [GECCO 2018], we formulated the simultaneous decision of the number and positions of wells as a mixed-integer problem solved by CMA-ES, and in [GECCO 2019], we sought placements whose performance is guaranteed over multiple geostatistically generated reservoir scenarios. In [ASC 2023], we proposed a method that searches while efficiently estimating the performance under the worst-case scenario, and applied it to well placement optimization. We have also developed an optimization tool that includes acceleration by parallel computing [JSCE 2021]. In October 2026, Taisei Corporation announced the well placement optimization technology for efficiently injecting CO₂ underground developed in this collaboration, and it was covered by Nikkei and Japanese construction industry newspapers (Taisei press release (Japanese), Nikkei (Japanese)).

関連する論文Related Publications

  1. GECCO 2018 ワークショップWorkshop口頭Oral
    Well Placement Optimization for Carbon dioxide Capture and Storage via CMA-ES with Mixed Integer Support
    Atsuhiro Miyagi, Youhei Akimoto, Hajime Yamamoto
    Proceedings of GECCO ’18 Companion, Kyoto, Japan, July 15-19, 2018, pp. 1696–1703 (2018)
  2. GECCO 2019 口頭Oral CORE A
    Well Placement Optimization under Geological Statistical Uncertainty
    Atsuhiro Miyagi, Youhei Akimoto, Hajime Yamamoto
    Proceedings of the Genetic and Evolutionary Computation Conference (GECCO ’19), July 13–17, 2019, Prague, Czech Republic, pp. 1284–1292 (2019)
  3. JSCE 2021 学生が筆頭Student-led
    二酸化炭素地中貯留における坑井配置の CMA-ES による最適化ツールの開発 ―並列計算手法による高速化―
    宮城 充宏, 山本 肇, 秋本 洋平
    土木学会論文集A2(応用力学), Vol.Vol. 77(1), pp. 21–34 (2021)
  4. ASC 2023 学生が筆頭Student-led JCR Q1
    Adaptive Scenario Subset Selection for Worst-Case Optimization and its Application to Well Placement Optimization
    Atsuhiro Miyagi, Kazuto Fukuchi, Jun Sakuma, Youhei Akimoto
    Applied Soft Computing, Vol.Vol. 133, Art. 109842 (2023)

PU 学習のための特徴量選択(日本総合研究所)Feature Selection for PU Learning (The Japan Research Institute)

キーワード:進化計算,特徴量選択,機械学習Keywords: Evolutionary Computation, Feature Selection, Machine Learning

日本総合研究所と,正例とラベルのないデータだけから学習する PU(Positive-Unlabeled)学習のための特徴量選択に取り組んでいます.異常検知のように,正例が少なく大部分のデータにラベルがない場面では,ラベルのないデータを負例とみなす通常の方法では,正例のデータの統計的な構造を見落とすおそれがあります.[GECCO 2025] では,PU 学習におけるクラスタ仮定に基づく目的関数を設計し,特徴量選択を 2 値の最適化問題として compact GA で解く方法を提案しました.この成果は日本総合研究所のニュースリリースでも紹介され,迷惑メールや不正送金の検知などへの応用が想定されています(日本総合研究所のニュースリリース).

With The Japan Research Institute, we study feature selection for PU (positive-unlabeled) learning, which learns only from positive and unlabeled data. In settings such as anomaly detection, where positives are few and most data are unlabeled, the usual approach of treating unlabeled data as negatives may miss the statistical structure of the positive data. In [GECCO 2025], we designed an objective based on the cluster assumption in PU learning and proposed a method that solves feature selection as a binary optimization problem with the compact GA. This result was also featured in a press release by The Japan Research Institute, which envisions applications such as detecting spam and fraudulent bank transfers (JRI press release (Japanese)).

関連する論文Related Publications

  1. GECCO 2025 口頭Oral CORE A
    Feature selection based on cluster assumption in PU learning
    Motonobu Uchikoshi, Youhei Akimoto
    GECCO '25: Proceedings of the Genetic and Evolutionary Computation Conference, pp. 490–498 (2025)

基礎研究での共同研究と学術指導Collaborations in Fundamental Research and Academic Consulting

基礎研究でも,国内外の研究者と共同研究を進めています.CMA-ES の考案者である Nikolaus Hansen 氏をはじめとする Inria(フランス)のグループとは,CMA-ES の設計と理論解析について長く共同研究を続けており,国際会議での CMA-ES のチュートリアルも共同で担当しています.ルール大学ボーフム(ドイツ)とは進化戦略の収束解析,横浜国立大学とはニューラルネットワークの構造探索や情報幾何的最適化,混合変数の最適化,LINEヤフーとは大規模言語モデルのアラインメントや安全な強化学習,二段階強化学習に取り組んでいます.このほか,企業名や内容は公開していませんが,企業への学術指導も行っています.

私達の技術が役立ちそうな興味深い応用をお持ちでしたら,お気軽にご連絡ください.

In fundamental research as well, we collaborate with researchers in Japan and abroad. With the group at Inria (France), including Nikolaus Hansen, the originator of CMA-ES, we have long collaborated on the design and theory of CMA-ES, and we jointly give CMA-ES tutorials at international conferences. We work with Ruhr University Bochum (Germany) on convergence analysis of evolution strategies, with Yokohama National University on neural architecture search, information-geometric optimization, and mixed-variable optimization, and with LY Corporation on alignment of large language models, safe reinforcement learning, and bilevel reinforcement learning. We also provide academic consulting to companies, whose names and topics are not disclosed.

If you have an interesting application where our techniques may help, please feel free to contact us.