• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to previous page

Development of Machine Learning Framework Based on Structure Optimization of Computational Graph

Research Project

Project/Area Number 20H04240
Research Category

Grant-in-Aid for Scientific Research (B)

Allocation TypeSingle-year Grants
Section一般
Review Section Basic Section 61030:Intelligent informatics-related
Research InstitutionYokohama National University

Principal Investigator

Shirakawa Shinichi  横浜国立大学, 大学院環境情報研究院, 准教授 (90633272)

Project Period (FY) 2020-04-01 – 2023-03-31
Project Status Completed (Fiscal Year 2022)
Budget Amount *help
¥17,160,000 (Direct Cost: ¥13,200,000、Indirect Cost: ¥3,960,000)
Fiscal Year 2022: ¥5,980,000 (Direct Cost: ¥4,600,000、Indirect Cost: ¥1,380,000)
Fiscal Year 2021: ¥5,980,000 (Direct Cost: ¥4,600,000、Indirect Cost: ¥1,380,000)
Fiscal Year 2020: ¥5,200,000 (Direct Cost: ¥4,000,000、Indirect Cost: ¥1,200,000)
Keywords機械学習 / 最適化 / 計算グラフ / 進化計算 / 知識獲得 / 構造最適化 / 勾配法
Outline of Research at the Start

現在の深層学習は,内部処理の解釈が困難,演算量やメモリ使用量が膨大でリソース効率が悪い,といった課題を抱えている.本研究では,処理内容が解釈可能な演算ユニットからなる計算グラフの構造を学習することで,これらの課題を解決する新しい機械学習フレームワークを開発する.開発する方式では,各演算ユニットの役割が解釈容易かつパラメータ数が削減されるため,内部処理の解釈性やリソース効率の改善が期待できる.さらに,構造学習の効率を高める学習方式を開発し,深層学習に匹敵する性能の実現をねらう.本研究は,深層学習を社会実装する際にボトルネックになっている課題の解決につながると考えられる.

Outline of Final Research Achievements

In this research, we developed machine learning methods for learning knowledge representation models from data by optimizing the structure of computational graphs. We showed that our methods could obtain interpretable and resource-efficient models by optimizing the computational graph consisting of interpretable operation units. In addition, to improve the scalability and efficiency of structure optimization, we developed and improved optimization algorithms based on the gradient methods using relaxation schemes called stochastic relaxation and continuous relaxation.

Academic Significance and Societal Importance of the Research Achievements

本研究で開発した計算グラフの構造最適化に基づく機械学習方式は、モデルの構造自体を効率的に学習できるという利点がある。これにより、解釈可能な演算ユニットからなる計算グラフの学習や、コンパクトな構造の学習が可能となるため、解釈性や計算効率の良いモデルが求められる応用で活用できる。さらに、本研究で開発した最適化方式は、計算グラフの構造最適化以外の問題にも応用できる可能性がある。

Report

(4 results)
  • 2022 Annual Research Report   Final Research Report ( PDF )
  • 2021 Annual Research Report
  • 2020 Annual Research Report
  • Research Products

    (18 results)

All 2023 2022 2021 2020

All Journal Article (9 results) (of which Peer Reviewed: 8 results) Presentation (9 results) (of which Int'l Joint Research: 1 results,  Invited: 1 results)

  • [Journal Article] Surrogate-Assisted (1+1)-CMA-ES with Switching Mechanism of Utility Functions2023

    • Author(s)
      Yutaro Yamada, Kento Uchida, Shota Saito, Shinichi Shirakawa
    • Journal Title

      Applications of Evolutionary Computation (EvoApplications 2023)

      Volume: 13989 of LNCS Pages: 798-814

    • DOI

      10.1007/978-3-031-30229-9_51

    • ISBN
      9783031302282, 9783031302299
    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] (1+1)-CMA-ES with Margin for Discrete and Mixed-Integer Problems2023

    • Author(s)
      Yohei Watanabe, Kento Uchida, Ryoki Hamano, Shota Saito, Masahiro Nomura, Shinichi Shirakawa
    • Journal Title

      Proceedings of the Genetic and Evolutionary Computation Conference (GECCO 2023)

      Volume: - Pages: 882-890

    • DOI

      10.1145/3583131.3590516

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] CMA-ES with Margin: Lower-Bounding Marginal Probability for Mixed-Integer Black-Box Optimization2022

    • Author(s)
      Ryoki Hamano, Shota Saito, Masahiro Nomura, Shinichi Shirakawa
    • Journal Title

      Proceedings of the Genetic and Evolutionary Computation Conference (GECCO 2022)

      Volume: - Pages: 639-647

    • DOI

      10.1145/3512290.3528827

    • Related Report
      2022 Annual Research Report 2021 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Reduction of Genetic Drift in Population-Based Incremental Learning via Entropy Regularization2022

    • Author(s)
      Ryoki Hamano, Shinichi Shirakawa
    • Journal Title

      Proceedings of the Genetic and Evolutionary Computation Conference Companion (GECCO 2022, Poster Paper)

      Volume: - Pages: 491-494

    • DOI

      10.1145/3520304.3529012

    • Related Report
      2022 Annual Research Report 2021 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Benchmarking CMA-ES with Margin on the bbob-mixint Testbed2022

    • Author(s)
      Ryoki Hamano, Shota Saito, Masahiro Nomura, Shinichi Shirakawa
    • Journal Title

      Proceedings of the Genetic and Evolutionary Computation Conference Companion (GECCO Workshop on Black-Box Optimization Benchmarking (BBOB 2022))

      Volume: - Pages: 1708-1716

    • DOI

      10.1145/3520304.3534043

    • Related Report
      2022 Annual Research Report 2021 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Efficient Search of Multiple Neural Architectures with Different Complexities via Importance Sampling2022

    • Author(s)
      Yuhei Noda, Shota Saito, Shinichi Shirakawa
    • Journal Title

      Proceedings of the 31st International Conference on Artificial Neural Networks (ICANN 2022)

      Volume: 13532 of LNCS Pages: 607-619

    • DOI

      10.1007/978-3-031-15937-4_51

    • ISBN
      9783031159367, 9783031159374
    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Improvement of sep-CMA-ES for Optimization of High-Dimensional Functions with Low Effective Dimensionality2022

    • Author(s)
      Teppei Yamaguchi, Kento Uchida, Shinichi Shirakawa
    • Journal Title

      Proceedings of the 2022 IEEE Symposium Series On Computational Intelligence (SSCI)

      Volume: - Pages: 1659-1668

    • DOI

      10.1109/ssci51031.2022.10022244

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] HACNet: End-to-end learning of table-to-image converter and convolutional neural network2022

    • Author(s)
      Takuya Matsuda, Kento Uchida, Shota Saito, Shinichi Shirakawa
    • Journal Title

      preprint

      Volume: -

    • DOI

      10.21203/rs.3.rs-2174672/v1

    • Related Report
      2022 Annual Research Report
  • [Journal Article] Adaptive Stochastic Natural Gradient Method for Optimizing Functions with Low Effective Dimensionality2020

    • Author(s)
      Teppei Yamaguchi, Kento Uchida, Shinichi Shirakawa
    • Journal Title

      Proceedings of the 16th International Conference on Parallel Problem Solving from Nature (PPSN XVI)

      Volume: 12269 Pages: 719-731

    • DOI

      10.1007/978-3-030-58112-1_50

    • ISBN
      9783030581114, 9783030581121
    • Related Report
      2020 Annual Research Report
    • Peer Reviewed
  • [Presentation] 構造の複雑さを考慮したNeural Architecture Searchにおける複数構造探索の効率化2022

    • Author(s)
      埜田 夕平,斉藤 翔汰,白川 真一
    • Organizer
      2022年度 人工知能学会全国大会 (第36回)
    • Related Report
      2022 Annual Research Report
  • [Presentation] 評価値への影響を持たない次元を含む高次元最適化問題のためのCMA-ESの改良2022

    • Author(s)
      内田 絢斗,山口 哲平,白川 真一
    • Organizer
      進化計算シンポジウム2022
    • Related Report
      2022 Annual Research Report
  • [Presentation] 利得関数の適応的切替機構を導入したサロゲートモデルを用いた(1+1)-CMA-ESの提案2022

    • Author(s)
      山田 裕太郎,内田 絢斗,斉藤 翔汰,白川 真一
    • Organizer
      進化計算シンポジウム2022
    • Related Report
      2022 Annual Research Report
  • [Presentation] 離散変数最適化および混合整数最適化のためのマージン補正付き(1+1)-CMA-ESの提案2022

    • Author(s)
      渡邉 陽平,内田 絢斗,濱野 椋希,斉藤 翔汰,野村 将寛,白川 真一
    • Organizer
      第23回 進化計算学会研究会
    • Related Report
      2022 Annual Research Report
  • [Presentation] Genetic Driftの抑制を目的とするエントロピー正則化を導入したPBILの提案2021

    • Author(s)
      濱野 椋希,白川 真一
    • Organizer
      進化計算シンポジウム2021
    • Related Report
      2021 Annual Research Report
  • [Presentation] 目的関数の単調増加変換に対する不変性をもつサロゲートモデルを用いた(1+1)-CMA-ESの提案2021

    • Author(s)
      山田 裕太郎,内田 絢斗,梅木 宏,山口 哲平,斉藤 翔汰,白川 真一
    • Organizer
      進化計算シンポジウム2021
    • Related Report
      2021 Annual Research Report
  • [Presentation] 遺伝的プログラミングを用いた船体運動モデル同定における入力端子の影響について2021

    • Author(s)
      巣山 凜,牧 敦生,宮内 新喜,白川 真一
    • Organizer
      令和3年 日本船舶海洋工学会 秋季講演会
    • Related Report
      2021 Annual Research Report
  • [Presentation] Evolutionary Machine Learning (Tutorial)2021

    • Author(s)
      Masaya Nakata, Will Browne, and Shinichi Shirakawa
    • Organizer
      2021 IEEE Congress on Evolutionary Computation (CEC 2021)
    • Related Report
      2021 Annual Research Report
    • Int'l Joint Research
  • [Presentation] 確率モデルに基づく進化計算とその応用2021

    • Author(s)
      白川真一
    • Organizer
      IMI研究集会「進化計算の数理」
    • Related Report
      2021 Annual Research Report
    • Invited

URL: 

Published: 2020-04-28   Modified: 2024-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi