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

Analysis for scalable Bayesian calculations

Research Project

Project/Area Number 21K18589
Research Category

Grant-in-Aid for Challenging Research (Exploratory)

Allocation TypeMulti-year Fund
Review Section Medium-sized Section 12:Analysis, applied mathematics, and related fields
Research InstitutionThe Institute of Statistical Mathematics

Principal Investigator

Kamatani Kengo  統計数理研究所, 統計基盤数理研究系, 教授 (00569767)

Project Period (FY) 2021-07-09 – 2024-03-31
Project Status Completed (Fiscal Year 2023)
Budget Amount *help
¥3,900,000 (Direct Cost: ¥3,000,000、Indirect Cost: ¥900,000)
Fiscal Year 2023: ¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000)
Fiscal Year 2022: ¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000)
Fiscal Year 2021: ¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000)
Keywordsマルコフ連鎖 / モンテカルロ法 / ベイズ統計学 / 確率過程 / ハミルトニアン / Markov chain / Monte Carlo / Bayesian statistics / Scalability / Bayesian Statistics / Exact sampling / Differential privacy / Scalable Computing / Monte Carlo method / Stochastic process / Random number generation / マルコフ過程 / スケーラブル / ビッグデータ
Outline of Research at the Start

ベイズ統計学においても,データサイズに関して頑健なアルゴリズムが注目を集めている.そうしたアルゴリズムは,正確で時間のかかる手法に取って代わり,いまでは基本的なアルゴリズムとみなされている.しかし,スケーラブルの実現のために,ベイズ統計学の特徴であった明瞭な意味を失ってしまう.
この数年に,従来の常識を覆す,明瞭な意味を保ついくつかの手法が提案されてきた.いずれも確率過程を用いた手法だ.確率過程の生成には技術的な困難がともなう.
本研究では技術的困難の解消の糸口を探りたい.また,そうした試みを通じて統計計算の発展に寄与したい.

Outline of Final Research Achievements

We are developing implementation techniques for a new robust method using piecewise deterministic Markov processes and studying its theoretical properties, with research still ongoing. Meanwhile, we also explored the Haar-Weave-Metropolis method, which can be considered a discrete-time version. Recently, many methods have used deterministic proposals based on local information but lack robustness. Conversely, existing robust methods are difficult to incorporate local information into. In our study, we developed the Haar-Weave-Metropolis kernel, which combines the strengths of both approaches, and demonstrated its superiority in terms of effective sample size and mean squared jump distance in numerical experiments.

Academic Significance and Societal Importance of the Research Achievements

本研究課題では、近年注目されている確率過程の生成を用いる手法の技術的課題であった確率過程の効率的な生成を目的としていた。国際共同研究により、技術的困難を軽減することができた。ベイズ統計学の分野では、マルコフ連鎖モンテカルロ法が依然として主流であり、この手法が苦手とする部分は、そのままベイズ統計学の適用が困難な部分でもあった。本研究の技術的な発展により、従来手法で困難とされていた領域をさらに狭めることができると期待している。

Report

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

    (12 results)

All 2023 2022 2021 Other

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

  • [Int'l Joint Research] Paris Dauphine University(フランス)

    • Related Report
      2023 Annual Research Report
  • [Int'l Joint Research] University of Warwick(英国)

    • Related Report
      2021 Research-status Report
  • [Int'l Joint Research] Delft University of Technology(オランダ)

    • Related Report
      2021 Research-status Report
  • [Journal Article] HAAR-WEAVE-METROPOLIS KERNEL2022

    • Author(s)
      KAMATANI Kengo、SONG Xiaolin
    • Journal Title

      Bulletin of informatics and cybernetics

      Volume: 54 Issue: 1 Pages: 1-31

    • DOI

      10.5109/4755997

    • NAID

      120007193399

    • ISSN
      0286-522X, 2435-743X
    • Related Report
      2021 Research-status Report
    • Peer Reviewed / Open Access
  • [Presentation] Haar-Weave-Metropolis Kernel2023

    • Author(s)
      Kengo Kamatani
    • Organizer
      ISI World Statistics Congress, 2023
    • Related Report
      2023 Annual Research Report
    • Int'l Joint Research / Invited
  • [Presentation] Scaling of Piecewise Deterministic Monte Carlo for Anisotropic Targets2023

    • Author(s)
      Kengo Kamatani
    • Organizer
      CMStatistics 2023
    • Related Report
      2023 Annual Research Report
    • Int'l Joint Research / Invited
  • [Presentation] Non-reversible guided Metropolis kernel2022

    • Author(s)
      Kengo Kamatani
    • Organizer
      Unification algorithmique d’analyses statistiques multiples Computational methods for unifying multiple statistical analyses (Fusion)
    • Related Report
      2022 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Scaling limit of Markov chain/process Monte Carlo methods2022

    • Author(s)
      Kengo Kamatani
    • Organizer
      IASC-ARS 2022
    • Related Report
      2021 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] High-dimensional asymptotics using multiscale analysis2021

    • Author(s)
      Kengo Kamatani
    • Organizer
      MCM 2021
    • Related Report
      2021 Research-status Report
    • Int'l Joint Research
  • [Remarks] https://sites.google.com/view/kengokamatani/home

    • Related Report
      2023 Annual Research Report
  • [Remarks] Kengo Kamatani

    • URL

      https://sites.google.com/view/kengokamatani/home?authuser=0

    • Related Report
      2022 Research-status Report
  • [Remarks] Kengo Kamatani

    • URL

      https://sites.google.com/view/kengokamatani/home

    • Related Report
      2021 Research-status Report

URL: 

Published: 2021-07-13   Modified: 2025-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi