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

Novel development in stochastic optimization using multilevel Monte Carlo methods

Research Project

Project/Area Number 20K03744
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 InstitutionThe University of Tokyo

Principal Investigator

Goda Takashi  東京大学, 大学院工学系研究科(工学部), 准教授 (50733648)

Project Period (FY) 2020-04-01 – 2023-03-31
Project Status Completed (Fiscal Year 2022)
Budget Amount *help
¥4,290,000 (Direct Cost: ¥3,300,000、Indirect Cost: ¥990,000)
Fiscal Year 2022: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Fiscal Year 2021: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Fiscal Year 2020: ¥2,210,000 (Direct Cost: ¥1,700,000、Indirect Cost: ¥510,000)
Keywordsモンテカルロ法 / マルチレベルモンテカルロ法 / 確率的最適化 / 確率的勾配降下法 / 入れ子型期待値 / ベイズ実験計画法 / ベイズ実験計画 / 確率的近似法 / 不偏推定
Outline of Research at the Start

確率的最適化とはパラメータを含む期待値を最小化あるいは最大化するための数値計算法であるが。既存のアルゴリズムでは期待値の(パラメータについての)勾配に対する不偏推定量を構成できることが陰に仮定されてきた。本研究では不偏推定量を構成することが必ずしも容易ではない、あるいは困難な状況に対する確率的最適化について考究する。マルチレベルモンテカルロ法を駆使した新奇的かつ明示的アルゴリズムを提示することによって、確率的最適化の適用可能性を拡張し、当該分野に新しい展開をもたらす。更に、ベイズ的実験計画法の最適化や変分オートエンコーダの学習といった具体的な問題へ応用し、その有効性を実験的にも示す。

Outline of Final Research Achievements

This research addressed the problem of maximizing or minimizing a quantity called nested expectation, which depends on some parameters. Such problem settings arise not only in optimizing Bayesian experimental designs but also in various scientific fields including machine learning. The aim of this research was to expand the applicability of stochastic gradient descent by constructing unbiased estimators for the gradient of the objective function, i.e., parametrized nested expectation. Specifically, we theoretically demonstrated that an unbiased gradient estimator with finite variance can be constructed by appropriately randomizing a multilevel Monte Carlo method, and we revealed the effectiveness of our novel approach through numerical experiments in various practical applications.

Academic Significance and Societal Importance of the Research Achievements

既存のアプローチとしては、入れ子になっている期待値のそれぞれをモンテカルロ法で近似するという方法が取られてきたが、どんなにサンプル数を増やしても不偏性を持たないことから、確率的最適化において正しい値に収束しないという問題点があった。本研究成果によってこの問題を解決し、より広いクラスの最適化問題に対して、確率的最適化が理論的にも実験的にも適切に最適解を探索できることになった点に学術的意義がある。また、様々な科学技術分野への応用が考えられるため、具体的な応用を通じて研究成果の社会的意義を見出すことが出来る。

Report

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

    (21 results)

All 2023 2022 2021 2020 Other

All Int'l Joint Research (5 results) Journal Article (11 results) (of which Int'l Joint Research: 4 results,  Peer Reviewed: 11 results,  Open Access: 1 results) Presentation (5 results) (of which Int'l Joint Research: 4 results,  Invited: 1 results)

  • [Int'l Joint Research] University of New South Wales(オーストラリア)

    • Related Report
      2022 Annual Research Report
  • [Int'l Joint Research] University of Montreal(カナダ)

    • Related Report
      2022 Annual Research Report
  • [Int'l Joint Research] University of Oxford(英国)

    • Related Report
      2021 Research-status Report
  • [Int'l Joint Research] University of New South Wales(オーストラリア)

    • Related Report
      2020 Research-status Report
  • [Int'l Joint Research] University of Oxford(英国)

    • Related Report
      2020 Research-status Report
  • [Journal Article] Constructing unbiased gradient estimators with finite variance for conditional stochastic optimization2023

    • Author(s)
      Takashi Goda Takashi, Wataru Kitade
    • Journal Title

      Mathematics and Computers in Simulation

      Volume: 204 Pages: 743-763

    • DOI

      10.1016/j.matcom.2022.09.012

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Improved bounds on the gain coefficients for digital nets in prime power base2023

    • Author(s)
      Takashi Goda, Kosuke Suzuki
    • Journal Title

      Journal of Complexity

      Volume: 76 Pages: 101722-101722

    • DOI

      10.1016/j.jco.2022.101722

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] An efficient estimation of nested expectations without conditional sampling2023

    • Author(s)
      Tomohiko Hironaka, Takashi Goda
    • Journal Title

      Journal of Computational and Applied Mathematics

      Volume: 421 Pages: 114811-114811

    • DOI

      10.1016/j.cam.2022.114811

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Polynomial tractability for integration in an unweighted function space with absolutely convergent Fourier series2023

    • Author(s)
      Takashi Goda
    • Journal Title

      Proceedings of the American Mathematical Society

      Volume: -

    • DOI

      10.1090/proc/16444

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Construction-Free Median Quasi-Monte Carlo Rules for Function Spaces with Unspecified Smoothness and General Weights2022

    • Author(s)
      Takashi Goda, Pierre L'Ecuyer
    • Journal Title

      SIAM Journal on Scientific Computing

      Volume: 44 Issue: 4 Pages: A2765-A2788

    • DOI

      10.1137/22m1473625

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Int'l Joint Research
  • [Journal Article] Component-by-component construction of randomized rank-1 lattice rules achieving almost the optimal randomized error rate2022

    • Author(s)
      Josef Dick, Takashi Goda, Kosuke Suzuki
    • Journal Title

      Mathematics of Computation

      Volume: 91 Pages: 2771-2801

    • DOI

      10.1090/mcom/3769

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Int'l Joint Research
  • [Journal Article] A note on concatenation of quasi-Monte Carlo and plain Monte Carlo rules in high dimensions2022

    • Author(s)
      Takashi Goda
    • Journal Title

      Journal of Complexity

      Volume: 72 Pages: 101647-101647

    • DOI

      10.1016/j.jco.2022.101647

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Unbiased MLMC stochastic gradient-based optimization of Bayesian experimental designs2022

    • Author(s)
      Takashi Goda, Tomohiko Hironaka, Wataru Kitade, Adam Foster
    • Journal Title

      SIAM Journal on Scientific Computing

      Volume: 44 Issue: 1 Pages: A286-A311

    • DOI

      10.1137/20m1338848

    • Related Report
      2021 Research-status Report
    • Peer Reviewed / Int'l Joint Research
  • [Journal Article] Probabilistic threshold analysis by pairwise stochastic approximation for decision-making under uncertainty2021

    • Author(s)
      Takashi Goda, Yuki Yamada
    • Journal Title

      Scientific Reports

      Volume: 11 Issue: 1 Pages: 19671-19671

    • DOI

      10.1038/s41598-021-99089-z

    • Related Report
      2021 Research-status Report
    • Peer Reviewed / Open Access
  • [Journal Article] A simple algorithm for global sensitivity analysis with Shapley effects2021

    • Author(s)
      Takashi Goda
    • Journal Title

      Reliability Engineering & System Safety

      Volume: 213 Pages: 107702-107702

    • DOI

      10.1016/j.ress.2021.107702

    • Related Report
      2021 Research-status Report
    • Peer Reviewed
  • [Journal Article] Toeplitz Monte Carlo2021

    • Author(s)
      Josef Dick, Takashi Goda, Hiroya Murata
    • Journal Title

      Statistics and Computing

      Volume: 31 Issue: 1 Pages: 1-15

    • DOI

      10.1007/s11222-020-09987-x

    • Related Report
      2020 Research-status Report
    • Peer Reviewed / Int'l Joint Research
  • [Presentation] Construction-free median lattice rules2022

    • Author(s)
      Takashi Goda
    • Organizer
      15th International Conference on Monte Carlo and Quasi-Monte Carlo Methods in Scientific Computing
    • Related Report
      2022 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Multilevel Monte Carlo methods for estimating the expected value of information2022

    • Author(s)
      Takashi Goda
    • Organizer
      SIAM Conference on Uncertainty Quantification (UQ22)
    • Related Report
      2022 Annual Research Report
    • Int'l Joint Research / Invited
  • [Presentation] Efficient debiased evidence estimation by multilevel Monte Carlo sampling2021

    • Author(s)
      Kei Ishikawa, Takashi Goda
    • Organizer
      Thirty-Seventh Conference on Uncertainty in Artificial Intelligence
    • Related Report
      2021 Research-status Report
    • Int'l Joint Research
  • [Presentation] Multilevel Monte Carlo methods for efficient nested simulations2021

    • Author(s)
      Takashi Goda
    • Organizer
      Satellite Bayesian/Monte Carlo workshop on Statistical modeling for stochastic processes and related fields
    • Related Report
      2021 Research-status Report
  • [Presentation] Two applications of multilevel Monte Carlo methods to Bayesian experimental designs2020

    • Author(s)
      Tomohiko Hironaka, Takashi Goda
    • Organizer
      14th International Conference on Monte Carlo and Quasi-Monte Carlo Methods in Scientific Computing (MCQMC 2020)
    • Related Report
      2020 Research-status Report
    • Int'l Joint Research

URL: 

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

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi