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

A Study on Controllable Representation Learning using Adversarial Training

Research Project

Project/Area Number 18K18101
Research Category

Grant-in-Aid for Early-Career Scientists

Allocation TypeMulti-year Fund
Review Section Basic Section 61030:Intelligent informatics-related
Research InstitutionThe University of Tokyo

Principal Investigator

Iwasawa Yusuke  東京大学, 大学院工学系研究科(工学部), 講師 (70808336)

Project Period (FY) 2018-04-01 – 2023-03-31
Project Status Completed (Fiscal Year 2022)
Budget Amount *help
¥4,030,000 (Direct Cost: ¥3,100,000、Indirect Cost: ¥930,000)
Fiscal Year 2021: ¥390,000 (Direct Cost: ¥300,000、Indirect Cost: ¥90,000)
Fiscal Year 2020: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Fiscal Year 2019: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Fiscal Year 2018: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000)
KeywordsDeep Learning / Firaness / Privacy / Transfer / Adversarial Training / 敵対的学習 / プライバシー保護 / ドメイン汎化 / 不変性 / 不変表現学習 / 深層学習 / 敵対的訓練 / プライバシー / フェアネス
Outline of Final Research Achievements

Throughout the research period, I achieved the following technical accomplishments:
(1) Analyzed the instability of the existing method, Adversarial Feature Learning, and proposed a solution (accepted in IJCAI2020 and other conferences). (2) Proposed a new criterion for invariance, called Sufficient Invariance, which maximizes invariance with respect to a factor of interest in an informationally novel range, and suggested methods to achieve Sufficient Invariance (accepted in ECML2019 and other conferences). (3) Proposed a framework based on graphical models to remove information from data without providing detailed information about the specific aspects the user wants to eliminate, and presented the corresponding methodology (accepted in ECML2021 and other conferences).

Academic Significance and Societal Importance of the Research Achievements

大規模言語モデルの登場などにより深層学習の実世界での活用は本格化しているが、通常の学習アルゴリズムは内部にあるバイアスを増長してしまう可能性がある。また、意図しない状況で不安定な挙動をすることがある。本研究の目的は、深層NNの表現が特定の情報を持たないように制御する要素技術の開発である。本研究成果により、未知ユーザの行動を高精度に認識したり、深層NNの判断基準が特定の因子によらないことを保証(プライバシー保護、公平性配慮)できると考えられる。

Report

(6 results)
  • 2022 Annual Research Report   Final Research Report ( PDF )
  • 2021 Research-status Report
  • 2020 Research-status Report
  • 2019 Research-status Report
  • 2018 Research-status Report
  • Research Products

    (11 results)

All 2021 2020 2019

All Journal Article (1 results) (of which Peer Reviewed: 1 results,  Open Access: 1 results) Presentation (10 results) (of which Int'l Joint Research: 6 results,  Invited: 1 results)

  • [Journal Article] Information-theoretic regularization for learning global features by sequential VAE2021

    • Author(s)
      Akuzawa Kei、Iwasawa Yusuke、Matsuo Yutaka
    • Journal Title

      Machine Learning

      Volume: 110 Issue: 8 Pages: 2239-2266

    • DOI

      10.1007/s10994-021-06032-4

    • Related Report
      2021 Research-status Report
    • Peer Reviewed / Open Access
  • [Presentation] Test-time classifier adjustment module for model-agnostic domain generalization2021

    • Author(s)
      Yusuke Iwasawa, Yutaka Matsuo
    • Organizer
      Advances in Neural Information Processing Systems
    • Related Report
      2021 Research-status Report
    • Int'l Joint Research
  • [Presentation] Stabilizing Adversarial Invariance Induction from Divergence Minimization Perspective2021

    • Author(s)
      Yusuke Iwasawa, Kei Akuzawa, Yutaka Matsuo
    • Organizer
      IJCAI
    • Related Report
      2020 Research-status Report
    • Int'l Joint Research
  • [Presentation] 制御可能な表現学習2020

    • Author(s)
      岩澤有祐
    • Organizer
      第5回 統計・機械学習若手シンポジウム
    • Related Report
      2020 Research-status Report
    • Invited
  • [Presentation] Stabilizing Adversarial Invariance Induction from Divergence Minimization Perspective2020

    • Author(s)
      Yusuke Iwasawa, Kei Akuzawa, Yutaka Matsuo
    • Organizer
      International Joint Conference of Artificial Intelligence (IJCAI)
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research
  • [Presentation] 分類性能による制約を考慮した敵対的不変表現学習によるドメイン汎化2019

    • Author(s)
      阿久澤 圭, 岩澤 有祐, 松尾 豊
    • Organizer
      人工知能学会全国大会
    • Related Report
      2019 Research-status Report 2018 Research-status Report
  • [Presentation] ペアワイズニューラルネット距離による不変表現学習2019

    • Author(s)
      岩澤 有祐, 阿久澤 圭, 松尾 豊
    • Organizer
      人工知能学会全国大会
    • Related Report
      2019 Research-status Report 2018 Research-status Report
  • [Presentation] 大域的な潜在変数を持つ系列変分自己符号化器による状態遷移モデルのメタ学習2019

    • Author(s)
      阿久澤圭, 岩澤有祐, 松尾豊,
    • Organizer
      情報論的学習理論ワークショップ
    • Related Report
      2019 Research-status Report
  • [Presentation] Adversarial Invariant Feature Learning with Accuracy Constraint for Domain Generalization2019

    • Author(s)
      Kei Akuzawa, Yusuke Iwasawa, and Yutaka Matsuo
    • Organizer
      ECMLPKDD
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research
  • [Presentation] Invariant Feature Learning by Attribute Perception Matching2019

    • Author(s)
      Yusuke Iwasawa, Kei Akuzawa, Yutaka Matsuo
    • Organizer
      International Conference of Learning Representations (Workshop)
    • Related Report
      2018 Research-status Report
    • Int'l Joint Research
  • [Presentation] Adversarial Feature Learning under Accuracy Constraint for Domain Generalization2019

    • Author(s)
      Kei Akuzawa, Yusuke Iwasawa, Yutaka Matsuo
    • Organizer
      International Conference of Learning Representations (Workshop)
    • Related Report
      2018 Research-status Report
    • Int'l Joint Research

URL: 

Published: 2018-04-23   Modified: 2024-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi