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

Study on deep neural nets with group theory

Research Project

Project/Area Number 20K03743
Research Category

Grant-in-Aid for Scientific Research (C)

Allocation TypeMulti-year Fund
Section一般
Review Section Basic Section 12040:Applied mathematics and statistics-related
Research InstitutionKyoto University (2023)
Institute of Physical and Chemical Research (2020-2022)

Principal Investigator

Sannai Akiyoshi  京都大学, 理学研究科, 特定准教授 (10610595)

Project Period (FY) 2020-04-01 – 2024-03-31
Project Status Completed (Fiscal Year 2023)
Budget Amount *help
¥4,420,000 (Direct Cost: ¥3,400,000、Indirect Cost: ¥1,020,000)
Fiscal Year 2022: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000)
Fiscal Year 2021: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000)
Fiscal Year 2020: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000)
Keywords深層学習 / 対称性 / 幾何学的深層学習 / グラフ理論 / 不変式論 / 群論 / メタ学習 / 構造的因果推論 / 深層ニューラルネット / 構造的因果モデル / 表現論
Outline of Research at the Start

本研究計画はZaheerらの定義した対称性を持った深層ニューラルネットの理論を群論、表現論、不変式論の見地から一般化、精密化するものである。一般化の最初のステップとしてSnの自然表現の二階のテンソル作用を考える。これは作用として置換作用の自然な一般化でありながら、グラフを入力とする関数を考える時に自然に現れる存在でもある。Zaheer の場合もそうであるように、構成のキーとなるのは作用空間の対称代数の不変式論である。またこの不変式環にはグラフ理論的な意味づけを与えることができ、この視点からグラフ理論的に自然な不変式環の生成元を探し、それを元に深層ニューラルネットを構成する。

Outline of Final Research Achievements

This research generalized and refined the theory of symmetric deep neural networks of Zaheer et al. from the viewpoints of group theory, representation theory, and invariant formula theory. Specifically, we considered the second-order tensor action of natural representations of Sn and achieved a generalization of functions with graphs as input. By using Reynolds operators, we found that it is possible to transform ordinary neural networks into a symmetric form and also to reduce the number of input variables. The results are expected to be used in the development of computationally efficient algorithms and in many application areas such as social network analysis. The research results have been published in JMLR.

Academic Significance and Societal Importance of the Research Achievements

本研究は、Zaheerらの対称性を持つ深層ニューラルネットワーク理論を群論、表現論、不変式論の視点から一般化し、精密化しました。学術的意義として、理論の拡張と深化、不変式論の応用、レイノルズ作用素の利用が挙げられます。これにより、深層学習モデルの設計に新たな視点が提供されました。社会的意義として、高効率なアルゴリズムの開発、グラフデータを扱う多分野での利用、技術の普及と教育の促進が期待されます。特に、ソーシャルネットワーク解析や交通ネットワーク解析などの応用分野での利用が進むことで、様々な社会課題の解決に寄与する可能性があります。

Report

(5 results)
  • 2023 Annual Research Report   Final Research Report ( PDF )
  • 2022 Research-status Report
  • 2021 Research-status Report
  • 2020 Research-status Report
  • Research Products

    (11 results)

All 2024 2021 2020 Other

All Int'l Joint Research (1 results) Journal Article (4 results) (of which Peer Reviewed: 4 results,  Open Access: 1 results) Presentation (5 results) (of which Int'l Joint Research: 3 results,  Invited: 1 results) Remarks (1 results)

  • [Int'l Joint Research] Ecole Normale Superieure Paris-Saclay(フランス)

    • Related Report
      2020 Research-status Report
  • [Journal Article] Invariant and Equivariant Reynolds Networks2024

    • Author(s)
      Akiyoshi Sannai, Makoto Kawano, Wataru Kumagai
    • Journal Title

      journal of machine learning research

      Volume: 25 Pages: 1-36

    • Related Report
      2023 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Group Equivariant Conditional Neural Processes2021

    • Author(s)
      Makoto Kawano, Wataru Kumagai, Akiyoshi Sannai, Yusuke Iwasawa, Yutaka Matsuo
    • Journal Title

      International Conference on Learning Representations

      Volume: -

    • Related Report
      2021 Research-status Report 2020 Research-status Report
    • Peer Reviewed
  • [Journal Article] On the number of linear functions composing deep neural network: Towards a refined definition of neural networks complexity2021

    • Author(s)
      Yuuki Takai, Akiyoshi Sannai, Matthieu Cordonnier
    • Journal Title

      The 24th International Conference on Artificial Intelligence and Statistics

      Volume: -

    • Related Report
      2021 Research-status Report 2020 Research-status Report
    • Peer Reviewed
  • [Journal Article] Improved Generalization Bounds of Group Invariant / Equivariant Deep Networks via Quotient Feature Spaces2021

    • Author(s)
      Akiyoshi Sannai, Masaaki Imaizumi, Makoto Kawano
    • Journal Title

      37th Conference on Uncertainty in Artificial Intelligence

      Volume: -

    • Related Report
      2021 Research-status Report
    • Peer Reviewed
  • [Presentation] Group Equivariant Conditional Neural Processes2021

    • Author(s)
      Makoto Kawano, Wataru Kumagai, Akiyoshi Sannai, Yusuke Iwasawa, Yutaka Matsuo
    • Organizer
      International Conference on Learning Representations
    • Related Report
      2021 Research-status Report 2020 Research-status Report
    • Int'l Joint Research
  • [Presentation] On the number of linear functions composing deep neural network: Towards a refined definition of neural networks complexity2021

    • Author(s)
      Yuuki Takai, Akiyoshi Sannai, Matthieu Cordonnier
    • Organizer
      The 24th International Conference on Artificial Intelligence and Statistics
    • Related Report
      2021 Research-status Report 2020 Research-status Report
    • Int'l Joint Research
  • [Presentation] Improved Generalization Bounds of Group Invariant / Equivariant Deep Networks via Quotient Feature Spaces2021

    • Author(s)
      Akiyoshi Sannai, Masaaki Imaizumi, Makoto Kawano
    • Organizer
      37th Conference on Uncertainty in Artificial Intelligence
    • Related Report
      2021 Research-status Report
    • Int'l Joint Research
  • [Presentation] 対称性を持つ深層学習2021

    • Author(s)
      三内顕義
    • Organizer
      東京大学大学院数理科学研究科情報数学セミナー
    • Related Report
      2020 Research-status Report
    • Invited
  • [Presentation] 群畳み込みニューラルネットワークによる同変的写像の普遍近似定理2020

    • Author(s)
      熊谷亘、三内顕義
    • Organizer
      統計関連学会連合大会
    • Related Report
      2020 Research-status Report
  • [Remarks] AISTATS2021

    • URL

      https://aistats.org/aistats2021/

    • Related Report
      2020 Research-status Report

URL: 

Published: 2020-04-28   Modified: 2025-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi