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

2018 Fiscal Year Annual Research Report

半解析リサンプリング法の開発と整備:信頼性評価への統計力学的アプローチ

Research Project

Project/Area Number 17H00764
Research InstitutionTokyo Institute of Technology

Principal Investigator

樺島 祥介  東京工業大学, 情報理工学院, 教授 (80260652)

Co-Investigator(Kenkyū-buntansha) 小渕 智之  東京工業大学, 情報理工学院, 助教 (40588448)
Project Period (FY) 2017-04-01 – 2022-03-31
Keywordsリサンプリング / 交差検証法 / ブートストラップ法 / レプリカ法 / 平均場近似
Outline of Annual Research Achievements

以下の結果を得た.
【課題1】交差検証法:多クラス分類問題に対する効率的な統計モデルである,L1正則化付き多クラス一般化線形モデルに対して,説明変数が独立で同一の分布から生成されていると仮定した場合に精密な評価であることが正当化される,1つ抜き交差検証誤差の半解析的評価法を与えた.またその成果を査読付き論文として公表した.Accelerating Cross-Validation in Multinomial Logistic Regression with L1-Regularization, Tomoyuki Obuchi, Yoshiyuki Kabashima, Journal of Machine Learning Research (2018) 19(52): (1-30)
【課題2】ブートストラップ法:L1正則化付き線形回帰モデルに対して,入出力対に対してリサンプリングを行うpair bootstrapping に関し,説明変数が独立に同一の分布から生成されていると仮定した場合に精密な評価であることが正当化される,推定量のブートストラップ平均/分散に関する半解析的評価法を与えた.現在,その成果をまとめて,査読付き学術誌に投稿中である.
【課題3】stability selection:【課題2】で開発したブートストラップ法に関する半解析的評価法にもとづいて,stability selection により,変数選択を行うための半解析的評価法を開発した.この方法では,ブートストラップによるデータ取得に関する統計的ゆらぎとともに正則化の強さに対するランダムなゆらぎを加えることで,重要変数の選択に関する安定性を評価する.【課題2】で述べた論文に含めて,現在,査読付き学術誌に投稿中である.

Current Status of Research Progress
Current Status of Research Progress

2: Research has progressed on the whole more than it was originally planned.

Reason

説明変数が独立で同一の分布から生成されていると仮定した場合について,【課題1】交差検証法,【課題2】ブートストラップ法,【課題3】:stability selection それぞれに対し,所望の統計量に関する半解析的評価法が得られている.これらは当初の計画に沿ったものであり概ね順調に進展していると考えられる.

Strategy for Future Research Activity

これまでは,理論的取扱が比較的容易な,説明変数が独立で同一の分布から生成されていると仮定した場合を中心に半解析的評価法の開発を行ってきた.しかしながら,実際のデータ解析ではこの仮定が妥当とは限らない.今後は,説明変数間に相関が含まれる場合にも対応できる公式の開発にも力を入れる.

  • Research Products

    (17 results)

All 2018 Other

All Int'l Joint Research (2 results) Journal Article (6 results) (of which Int'l Joint Research: 1 results,  Peer Reviewed: 6 results,  Open Access: 5 results) Presentation (9 results) (of which Int'l Joint Research: 9 results,  Invited: 5 results)

  • [Int'l Joint Research] Aalto University(フィンランド)

    • Country Name
      FINLAND
    • Counterpart Institution
      Aalto University
  • [Int'l Joint Research] University of Oslo(ノルウェー)

    • Country Name
      NORWAY
    • Counterpart Institution
      University of Oslo
  • [Journal Article] Statistical mechanical analysis of sparse linear regression as a variable selection problem2018

    • Author(s)
      Tomoyuki Obuchi, Yoshinori Nakanishi-Ohno, Masato Okada and Yoshiyuki Kabashima
    • Journal Title

      Journal of Statistical Mechanics: Theory and Experiment

      Volume: 2018 Pages: 103401(1-41)

    • DOI

      https://doi.org/10.1088/1742-5468/aae02c

    • Peer Reviewed / Open Access
  • [Journal Article] Accelerating Cross-Validation in Multinomial Logistic Regression with l1-Regularization2018

    • Author(s)
      Tomoyuki Obuchi and Yoshiyuki Kabashima
    • Journal Title

      The Journal of Machine Learning Research

      Volume: 19 Pages: 1-30

    • DOI

      http://www.jmlr.org/papers/volume19/17-684/17-684.pdf

    • Peer Reviewed / Open Access
  • [Journal Article] A statistical mechanics approach to de-biasing and uncertainty estimation in LASSO for random measurements2018

    • Author(s)
      Takahashi Takashi、Kabashima Yoshiyuki
    • Journal Title

      Journal of Statistical Mechanics: Theory and Experiment

      Volume: 2018 Pages: 073405(1-25)

    • DOI

      https://doi.org/10.1088/1742-5468/aace2e

    • Peer Reviewed / Open Access
  • [Journal Article] Inverse finite-size scaling for high-dimensional significance analysis2018

    • Author(s)
      Xu Yingying、Puranen Santeri、Corander Jukka、Kabashima Yoshiyuki
    • Journal Title

      Physical Review E

      Volume: 97 Pages: 062112(1-9)

    • DOI

      https://doi.org/10.1103/PhysRevE.97.062112

    • Peer Reviewed / Int'l Joint Research
  • [Journal Article] Objective and efficient inference for couplings in neuronal networks2018

    • Author(s)
      Yu Terada, Tomoyuki Obuchi, Takuya Isomura, Yoshiyuki Kabashima
    • Journal Title

      Advances in Neural Information Processing Systems

      Volume: 31 Pages: 4971-4980

    • DOI

      http://papers.nips.cc/paper/7745-objective-and-efficient-inference-for-couplings-in-neuronal-networks.pdf

    • Peer Reviewed / Open Access
  • [Journal Article] Mean-field theory of graph neural networks in graph partitioning2018

    • Author(s)
      Tatsuro Kawamoto, Masashi Tsubaki, Tomoyuki Obuchi
    • Journal Title

      Advances in Neural Information Processing Systems

      Volume: 31 Pages: 4361-4371

    • DOI

      http://papers.nips.cc/paper/7689-mean-field-theory-of-graph-neural-networks-in-graph-partitioning.pdf

    • Peer Reviewed / Open Access
  • [Presentation] Objective and efficient inference for couplings in neuronal networks2018

    • Author(s)
      Yu Terada, Tomoyuki Obuchi, Takuya Isomura, Yoshiyuki Kabashima
    • Organizer
      Thirty-second Conference on Neural Information Processing Systems (NeurIPS 2018)
    • Int'l Joint Research
  • [Presentation] Mean-field theory of graph neural networks in graph partitioning2018

    • Author(s)
      Tatsuro Kawamoto, Masashi Tsubaki, Tomoyuki Obuchi
    • Organizer
      Thirty-second Conference on Neural Information Processing Systems (NeurIPS 2018)
    • Int'l Joint Research
  • [Presentation] Accelerating Cross-Validation in Multinomial Logistic Regression with L1-Regularization2018

    • Author(s)
      Tomoyuki Obuchi and Yoshiyuki Kabashima
    • Organizer
      Statistical Physics and Machine Learning Back Together
    • Int'l Joint Research
  • [Presentation] A statistical mechanics approach to de-biasing and uncertainty estimation in LASSO2018

    • Author(s)
      Takashi Takahashi and Yoshiyuki Kabashima
    • Organizer
      Statistical Physics and Machine Learning Back Together
    • Int'l Joint Research / Invited
  • [Presentation] Statistical mechanical analysis of sparse linear regression as a variable selection problem2018

    • Author(s)
      Tomoyuki Obuchi, Y. Nakanishi-Ohno, M. Okada and Y. Kabashima
    • Organizer
      Disordered serendipity: a glassy path to discovery
    • Int'l Joint Research
  • [Presentation] Semi-analytic resampling in LASSO2018

    • Author(s)
      Yoshiyuki Kabashima
    • Organizer
      Disordered serendipity: a glassy path to discovery
    • Int'l Joint Research / Invited
  • [Presentation] Semi-analytic resampling in LASSO2018

    • Author(s)
      Yoshiyuki Kabashima
    • Organizer
      International Symposium on Turbo Codes & Iterative Information Processing 2018
    • Int'l Joint Research / Invited
  • [Presentation] Replica computation of perceptrons2018

    • Author(s)
      Yoshiyuki Kabashima
    • Organizer
      Physics, Inference and Learning
    • Int'l Joint Research / Invited
  • [Presentation] Objective and efficient inference for couplings in neuronal networks2018

    • Author(s)
      Yu Terada, Tomoyuki Obuchi, Takuya Isomura, Yoshiyuki Kabashima
    • Organizer
      Physics, Inference and Learning
    • Int'l Joint Research / Invited

URL: 

Published: 2019-12-27  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi