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

Development of Function Estimation Theory to Investigate the Principles of Deep Learning

Research Project

Project/Area Number 18K18114
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 (2020-2021)
The Institute of Statistical Mathematics (2018-2019)

Principal Investigator

Imaizumi Masaaki  東京大学, 大学院総合文化研究科, 准教授 (90814088)

Project Period (FY) 2018-04-01 – 2022-03-31
Project Status Discontinued (Fiscal Year 2021)
Budget Amount *help
¥3,900,000 (Direct Cost: ¥3,000,000、Indirect Cost: ¥900,000)
Fiscal Year 2021: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
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: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
Keywords深層学習 / 関数推定 / ノンパラメトリック統計 / 汎化誤差 / ノンパラメトリック統計学 / 学習理論 / ミニマックス最適性 / 機械学習
Outline of Final Research Achievements

The outline of this research plan is to construct a theory that can explain the principles of deep learning. Although deep learning has demonstrated high performance in practical applications, its principles are still largely unexplained, and a theory that can explain this performance is still under development.
In this research project, we have achieved the following results: (i) We proved that deep learning is superior when the true function generating the data has special properties such as singularity. (ii) Developed a theory showing that complex non-convex loss functions in deep learning can avoid overlearning. (iii) We showed that double-descent phenomenon under over-parameterization also occurs in deep learning models. (iv) Developed an algorithm to solve nonconvex optimization problems using the theoretical findings.

Academic Significance and Societal Importance of the Research Achievements

深層学習は、その高い性能から強い注目を集め、社会の各所で応用されている重要な技術である。今回のAIブームにおいて、深層学習によって実現する技術は数多い。
しかしながら、深層学習の原理は未だ十分に解明されていないのが現状である。その結果として、深層学習の欠点である膨大な計算コストや、ブラックボックスな挙動などの問題点は、未解決のまま残っている。これらの問題を根本から解決するには、基礎研究を通じて深層学習の原理を理解し、抜本的な解決手法を開発することが望まれる。
本研究はその試みの一端として、深層学習という新しい技術を数学的に記述することを試み、そして深層学習の成功の要因を明らかにする理論を構築した。

Report

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

    (85 results)

All 2021 2020 2019 2018 Other

All Int'l Joint Research (10 results) Journal Article (8 results) (of which Int'l Joint Research: 2 results,  Peer Reviewed: 7 results,  Open Access: 7 results) Presentation (65 results) (of which Int'l Joint Research: 25 results,  Invited: 41 results) Remarks (1 results) Funded Workshop (1 results)

  • [Int'l Joint Research] ラトガース大学/コーネル大学(米国)

    • Related Report
      2020 Annual Research Report
  • [Int'l Joint Research] インド工科大学(インド)

    • Related Report
      2020 Annual Research Report
  • [Int'l Joint Research] マックスプランク研究所(ドイツ)

    • Related Report
      2020 Annual Research Report
  • [Int'l Joint Research] ライデン大学(オランダ)

    • Related Report
      2020 Annual Research Report
  • [Int'l Joint Research] ロンドン経済学校(英国)

    • Related Report
      2020 Annual Research Report
  • [Int'l Joint Research] インド工科大学(インド)

    • Related Report
      2019 Research-status Report
  • [Int'l Joint Research] ペンシルバニア州立大学/コーネル大学/ニューヨーク大学(米国)

    • Related Report
      2019 Research-status Report
  • [Int'l Joint Research] ライデン大学(オランダ)

    • Related Report
      2019 Research-status Report
  • [Int'l Joint Research] Cornell University(米国)

    • Related Report
      2018 Research-status Report
  • [Int'l Joint Research] Leiden University(オランダ)

    • Related Report
      2018 Research-status Report
  • [Journal Article] 深層学習の原理解析:汎化誤差の側面から2021

    • Author(s)
      今泉允聡
    • Journal Title

      日本統計学会誌

      Volume: 50 Pages: 257-283

    • NAID

      130007995094

    • Related Report
      2020 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] On Random Subsampling of Gaussian Process Regression2020

    • Author(s)
      Kohei.Hayashi. Masaaki.Imaizumi. Yuichi Yoshida
    • Journal Title

      PMLR: Artificial Intelligence and Statistics

      Volume: 108 Pages: 2055-2065

    • Related Report
      2020 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Adaptive Approximation and Generalization of Deep Neural Network with Intrinsic Dimensionality2020

    • Author(s)
      Ryumei Nakada. Masaaki Imaizumi
    • Journal Title

      Journal of Machine Learning Research

      Volume: 21 Pages: 1-38

    • Related Report
      2020 Annual Research Report
    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] On Random Subsampling of Gaussian Process Regression: A Graphon-Based Analysis2020

    • Author(s)
      K.Hayashi, M.Imaizumi, Y,Yoshida
    • Journal Title

      Proceedings of Machine Learning Research (AI & Statistics)

      Volume: To appear

    • Related Report
      2019 Research-status Report
    • Peer Reviewed / Open Access
  • [Journal Article] Deep Neural Networks Learn Non-Smooth Functions Effectively2019

    • Author(s)
      M.Imaizumi, K.Fukumizu
    • Journal Title

      Proceedings of Machine Learning Research (AI & Statistics)

      Volume: 84 Pages: 869-878

    • Related Report
      2019 Research-status Report
    • Peer Reviewed / Open Access
  • [Journal Article] A simple method to construct confidence bands in functional linear regression2019

    • Author(s)
      Masaaki Imaizumi, Kengo Kato
    • Journal Title

      Statistica Sinica

      Volume: 印刷中

    • Related Report
      2018 Research-status Report
    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Statistically Efficient Estimation for Non-Smooth Probability Densities2018

    • Author(s)
      Masaaki Imazumi, Takanori Maehara, Yuichi Yoshida
    • Journal Title

      Proceedings of Machine Learning Research Workshop & Conference Proceedings (AISTATS 2018)

      Volume: 84 Pages: 978-987

    • Related Report
      2018 Research-status Report
    • Peer Reviewed / Open Access
  • [Journal Article] Deep Neural Networks Learn Non-Smooth Functions Effectively2018

    • Author(s)
      Masaaki Imaizumi, Kenji Fukumizu
    • Journal Title

      2018年度統計関連学会連合大会講演予稿集

      Volume: -

    • Related Report
      2018 Research-status Report
  • [Presentation] Generalization Analysis of Deep Models with Loss Surface and Over Parameterization2021

    • Author(s)
      Masaaki Imaizumiツ?
    • Organizer
      Workshop on Functional Inference and Machine Intelligence
    • Related Report
      2020 Annual Research Report
    • Int'l Joint Research / Invited
  • [Presentation] 深層学習の原理を明らかにする理論の試み2020

    • Author(s)
      今泉允聡
    • Organizer
      九大IMI特別セミナー
    • Related Report
      2020 Annual Research Report
    • Invited
  • [Presentation] 深層学習の原理を明らかにする理論の試み2020

    • Author(s)
      今泉允聡
    • Organizer
      ビッグデータCREST合宿
    • Related Report
      2020 Annual Research Report
    • Invited
  • [Presentation] Statistical Inference on M-estimators by High-dimensional Gaussian Approximation2020

    • Author(s)
      今泉允聡
    • Organizer
      東北大学データサイエンスセミナー
    • Related Report
      2020 Annual Research Report
    • Invited
  • [Presentation] Generalization Analysis for Mechanism of Deep Learning2020

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Seoul National University Seminar
    • Related Report
      2020 Annual Research Report
    • Int'l Joint Research / Invited
  • [Presentation] Statistical inference on M-estimators by high-dimensional Gaussian approximation2020

    • Author(s)
      Masaaki Imaizumiツ?
    • Organizer
      Workshop on Functional Inference and Machine Intelligence
    • Related Report
      2020 Annual Research Report
    • Int'l Joint Research / Invited
  • [Presentation] 深層学習の原理を明らかにする理論の試み2020

    • Author(s)
      今泉允聡
    • Organizer
      大阪大学MMDS AI・データ利活用研究会
    • Related Report
      2020 Annual Research Report
    • Invited
  • [Presentation] 深層学習の原理を明らかにする理論の試み2020

    • Author(s)
      今泉允聡
    • Organizer
      MLSE夏合宿
    • Related Report
      2020 Annual Research Report
    • Invited
  • [Presentation] 深層学習の理論2020

    • Author(s)
      今泉允聡
    • Organizer
      東大松尾研
    • Related Report
      2020 Annual Research Report
    • Invited
  • [Presentation] 深層学習の原理の理解に向けた理論の試み2020

    • Author(s)
      今泉允聡
    • Organizer
      諸科学における大規模データと統計数理モデリング
    • Related Report
      2020 Annual Research Report
    • Invited
  • [Presentation] On Random Subsampling of Gaussian Process Regression2020

    • Author(s)
      K.Hayashi. M.Imaizumi. Y.Yoshida
    • Organizer
      Artificial Intelligence and Statistics
    • Related Report
      2020 Annual Research Report
    • Int'l Joint Research
  • [Presentation] 大規模モデルの複雑性尺度と情報量規準2020

    • Author(s)
      今泉允聡
    • Organizer
      統計関連学会連合大会
    • Related Report
      2020 Annual Research Report
  • [Presentation] 損失形状に基づく深層学習の汎化誤差解析2020

    • Author(s)
      今泉允聡
    • Organizer
      統計関連学会連合大会
    • Related Report
      2020 Annual Research Report
  • [Presentation] 深層学習の発見からもたらされる基礎理論のパラダイム2020

    • Author(s)
      今泉允聡
    • Organizer
      東大情報理工コンピュータ専攻講演会
    • Related Report
      2020 Annual Research Report
    • Invited
  • [Presentation] 損失関数の形状に基づく深層学習の汎化誤差解析2020

    • Author(s)
      今泉允聡
    • Organizer
      RAMP(OR学会分科会)
    • Related Report
      2020 Annual Research Report
    • Invited
  • [Presentation] Theory of Deep Learning2020

    • Author(s)
      今泉允聡
    • Organizer
      奈良先端科学技術大学 データサイエンス特別講義
    • Related Report
      2020 Annual Research Report
    • Invited
  • [Presentation] Fast Convergence on Perfect Classification for Functional Data2020

    • Author(s)
      Tomoya Wakayama. Masaaki Imaizumi
    • Organizer
      CM-Statistics
    • Related Report
      2020 Annual Research Report
    • Int'l Joint Research / Invited
  • [Presentation] Statistical inference on M-estimators by high-dimensional Gaussian approximation2020

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Workshop on Functional Inference and Machine Intelligence 2020
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] 層学習の原理を明らかにする理論の試み2019

    • Author(s)
      今泉允聡
    • Organizer
      統計数理研究所オープンハウス
    • Related Report
      2019 Research-status Report
    • Invited
  • [Presentation] 深層学習による関数推定と特異性2019

    • Author(s)
      今泉允聡
    • Organizer
      京都大学RIMS共同研究集会
    • Related Report
      2019 Research-status Report
    • Invited
  • [Presentation] 深層学習の理論2019

    • Author(s)
      今泉允聡
    • Organizer
      IBIS2020 企画セッション
    • Related Report
      2019 Research-status Report
    • Invited
  • [Presentation] 深層学習入門2019

    • Author(s)
      今泉允聡
    • Organizer
      IBIS2020 チュートリアル
    • Related Report
      2019 Research-status Report
    • Invited
  • [Presentation] 深層学習の原理を明らかにする理論の試み2019

    • Author(s)
      今泉允聡
    • Organizer
      ビッグデータCREST合宿
    • Related Report
      2019 Research-status Report
    • Invited
  • [Presentation] 深層学習の原理を明らかにする理論の試み2019

    • Author(s)
      今泉允聡
    • Organizer
      大阪大学MMDS AI・データ利活用研究会
    • Related Report
      2019 Research-status Report
    • Invited
  • [Presentation] Deep Neural Networks Learn Non-Smooth Functions Effectively2019

    • Author(s)
      今泉允聡
    • Organizer
      AIMaP若手数学者交流会
    • Related Report
      2019 Research-status Report
  • [Presentation] Statistical Inference on M-estimators by High-dimensional Gaussian Approximation2019

    • Author(s)
      今泉允聡
    • Organizer
      日本統計学会春季集会
    • Related Report
      2019 Research-status Report
  • [Presentation] Generalization Analysis for Mechanism of Deep Neural Networks via Nonparametric Statistics2019

    • Author(s)
      今泉允聡
    • Organizer
      理研AIP数学セミナー
    • Related Report
      2019 Research-status Report
  • [Presentation] 深層学習の原理を明らかにする理論の試み2019

    • Author(s)
      今泉允聡
    • Organizer
      松尾研セミナー
    • Related Report
      2019 Research-status Report
  • [Presentation] 深層学習の高速化にむけた適応ネットワークの数学的発見と学習法開発2019

    • Author(s)
      今泉允聡
    • Organizer
      さきがけ領域会議
    • Related Report
      2019 Research-status Report
  • [Presentation] 深層学習・関数データ・高次元ガウス近似2019

    • Author(s)
      今泉允聡
    • Organizer
      Overfit Summer Seminar
    • Related Report
      2019 Research-status Report
  • [Presentation] 深層学習の高速化にむけた適応ネットワークの数学的発見と学習法開発2019

    • Author(s)
      今泉允聡
    • Organizer
      情報系WinterFesta
    • Related Report
      2019 Research-status Report
  • [Presentation] 深層学習の原理を明らかにする理論の試み2019

    • Author(s)
      今泉允聡
    • Organizer
      九大IMI特別セミナー
    • Related Report
      2019 Research-status Report
  • [Presentation] Statistical Inference on M-estimators by High-dimensional Gaussian Approximation2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      東北大学データサイエンスセミナー
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research
  • [Presentation] 深層学習の高速化にむけた適応ネットワークの数学的発見と学習法開発2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      さきがけ領域会議
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research
  • [Presentation] Generalization Analysis for Mechanism of Deep Learning via Nonparametric Statistics2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      International Chinese Statistical Association
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Statistical inference on M-estimators by high-dimensional Gaussian approximation2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      CM-Statistics
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Generalization Analysis for Mechanism of Deep Learning via Nonparametric Statistics2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      International Statistical Institute World Statistics Congress 2019
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Statistical Estimation for Non-Smooth Functions by Deep Neural Networks2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Joint Statistical Meeting
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Generalization Analysis for Mechanism of Deep Learning via Nonparametric Statistics2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Third International Workshop on Symbolic-Neural Learning (SNL-2019)
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Deep Neural Networks Learn Non-Smooth Functions Effectively2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      IMS-China International Conference on Statistics and Probability
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Inference on level sets in functional linear regression2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Econometrics and Statistics
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Generalization Analysis for Mechanism of Deep Learning by Statistics and Learning Theory2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Seoul National University Seminar
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research
  • [Presentation] Generalization Analysis for Mechanism of Deep Learning by Statistics and Learning Theory2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Penn State University Seminar
    • Related Report
      2019 Research-status Report
    • Int'l Joint Research
  • [Presentation] 統計理論によるGANの原理解析2019

    • Author(s)
      今泉 允聡
    • Organizer
      Preferred Networks セミナー
    • Related Report
      2018 Research-status Report
    • Invited
  • [Presentation] 高次元ガウス近似によるM推定量の統計的推論2019

    • Author(s)
      今泉 允聡
    • Organizer
      日本統計学会春季集会
    • Related Report
      2018 Research-status Report
  • [Presentation] Statistical Analysis for Generative Adversarial Networks2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Workshop on Functional Inference and Machine Intelligence
    • Related Report
      2018 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Analysis for Deep Learning by Function Estimation Theory2019

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      ISI-ISM-ISSAS Joint Conference 2019
    • Related Report
      2018 Research-status Report
    • Int'l Joint Research
  • [Presentation] Statistically Efficient Estimation for Non-Smooth Probability Densities2018

    • Author(s)
      今泉 允聡
    • Organizer
      The 21st International Conference on Artificial Intelligence and Statistics
    • Related Report
      2018 Research-status Report
    • Int'l Joint Research
  • [Presentation] Deep Neural Networks Learn Non-Smooth Functions Effectively2018

    • Author(s)
      今泉 允聡
    • Organizer
      Deep Neural Networks Learn Non-Smooth Functions Effectively
    • Related Report
      2018 Research-status Report
    • Invited
  • [Presentation] Deep Neural Networks Learn Non-Smooth Functions Effectively2018

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      ICML 2018 Workshop on Theory of Deep Learning
    • Related Report
      2018 Research-status Report
    • Int'l Joint Research
  • [Presentation] Deep Neural Networks Learn Non-Smooth Functions Effectively2018

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      統計関連学会連合大会
    • Related Report
      2018 Research-status Report
  • [Presentation] 汎化誤差評価によるGANの理論解析2018

    • Author(s)
      今泉 允聡
    • Organizer
      IBIS2018
    • Related Report
      2018 Research-status Report
  • [Presentation] 関数推定の理論による深層学習の原理解析2018

    • Author(s)
      今泉 允聡
    • Organizer
      RIMS共同研究「高度情報化社会に向けた数理最適化の新潮流」
    • Related Report
      2018 Research-status Report
    • Invited
  • [Presentation] 関数推定の理論による深層学習の原理解析2018

    • Author(s)
      今泉 允聡
    • Organizer
      統計・機械学習若手シンポジウム「統計・機械学習の交わりと拡がり」
    • Related Report
      2018 Research-status Report
    • Invited
  • [Presentation] 関数推定の理論による深層学習の原理解析2018

    • Author(s)
      今泉 允聡
    • Organizer
      応用数理学会
    • Related Report
      2018 Research-status Report
    • Invited
  • [Presentation] 関数推定の理論による深層学習の原理解析2018

    • Author(s)
      今泉 允聡
    • Organizer
      生命医薬情報学連合大会
    • Related Report
      2018 Research-status Report
    • Invited
  • [Presentation] 非滑らかな確率密度推定の統計理論的解析2018

    • Author(s)
      今泉 允聡
    • Organizer
      IBIS2018
    • Related Report
      2018 Research-status Report
    • Invited
  • [Presentation] 深層学習の概要とその理論研究の現状について2018

    • Author(s)
      今泉 允聡
    • Organizer
      SICE制御部門 データ科学とリンクした次世代の適応学習制御調査研究会
    • Related Report
      2018 Research-status Report
    • Invited
  • [Presentation] Szemeredi分割による非滑らかな密度関数の推定2018

    • Author(s)
      今泉 允聡
    • Organizer
      統計数理セミナー
    • Related Report
      2018 Research-status Report
  • [Presentation] 関数データ回帰の信頼バンド構成法2018

    • Author(s)
      今泉 允聡
    • Organizer
      Overfit Summer Seminar
    • Related Report
      2018 Research-status Report
    • Invited
  • [Presentation] 関数推定の理論による深層学習の原理解析2018

    • Author(s)
      今泉 允聡
    • Organizer
      数学と諸分野の連携にむけた若手数学者交流会
    • Related Report
      2018 Research-status Report
  • [Presentation] A simple method to construct confidence bands in functional linear regression2018

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Econometrics and Statistics 2018
    • Related Report
      2018 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Statistical Estimation for Non-Smooth Functions with the Regularity Lemma2018

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Discrete Optimization and Machine Learning Workshop
    • Related Report
      2018 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Inference on active domains of functional data via functional linear regression2018

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Computational and Methodological Statistics
    • Related Report
      2018 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Deep Neural Networks Learn Non-Smooth Functions Effectively2018

    • Author(s)
      Masaaki Imaizumi
    • Organizer
      Statistics Seminar, Mathematical Institute, Leiden University
    • Related Report
      2018 Research-status Report
    • Int'l Joint Research
  • [Remarks] 深層学習によるデータ固有のフラクタル構造などへの適応を証明

    • URL

      https://www.u-tokyo.ac.jp/focus/ja/articles/z0508_00102.html

    • Related Report
      2020 Annual Research Report
  • [Funded Workshop] Workshop on Functional Inference and Machine Intelligence2020

    • Related Report
      2020 Annual Research Report

URL: 

Published: 2018-04-23   Modified: 2023-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi