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

Meta-modeling method for a set of energy-based models

Research Project

Project/Area Number 22K17951
Research Category

Grant-in-Aid for Early-Career Scientists

Allocation TypeMulti-year Fund
Review Section Basic Section 61030:Intelligent informatics-related
Research InstitutionKyushu Institute of Technology

Principal Investigator

Hideaki Ishibashi  九州工業大学, 大学院生命体工学研究科, 助教 (30838389)

Project Period (FY) 2022-04-01 – 2025-03-31
Project Status Completed (Fiscal Year 2024)
Budget Amount *help
¥2,340,000 (Direct Cost: ¥1,800,000、Indirect Cost: ¥540,000)
Fiscal Year 2023: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
Fiscal Year 2022: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000)
KeywordsEnergy Based Model / スコアマッチング / メタ学習 / マルチタスク学習 / メタモデリング / 指数型分布族 / Energy-based model / 情報幾何学
Outline of Research at the Start

本研究では複数の類似する経験を通して得られた知識からそれらの知識に共通するより普遍的な知識を推定するメタモデリング法の学習理論を構築することを目指す.本研究の特色はEnergy-based model (EBM)集合のメタモデリング法の理論を構築することで,それぞれの経験を知識としてモデル化するモデリング法に依存しないユニバーサルなメタモデリングの学習理論構築を目指す点である.これにより既存のモデリング法のメタモデリングができるようになるだけでなく,データに合わせてシームレスに適切なメタモデリングを選択できるようになる.

Outline of Final Research Achievements

The contributions of this study are as follows:(1) We developed a meta-modeling method for a set of arbitrary exponential families using score matching. (2) We extended the method to handle a set of infinite-dimensional exponential families. (3) We proposed a meta-modeling method for a set of pretrained neural networks using denoising score matching, aimed at meta-modeling for deep learning. (4) We showed that the resulting meta-model belongs to the exponential family, thereby enabling to define an adaptive exponential families.

Academic Significance and Societal Importance of the Research Achievements

新しい環境や未知の現象に少数ショットで即座に対応することは実環境下で動く機械学習を実用化する上で重要な要素である.本研究ではこのような課題に対し,それぞれの環境や現象ごとにスコアマッチングと呼ばれる方法で学習したモデルをさらにモデル化するメタモデルを推定する方法を開発し,さまざまな問題に広く使える汎用的な手法を実現した.

Report

(4 results)
  • 2024 Annual Research Report   Final Research Report ( PDF )
  • 2023 Research-status Report
  • 2022 Research-status Report
  • Research Products

    (9 results)

All 2025 2024 2023

All Journal Article (1 results) (of which Peer Reviewed: 1 results,  Open Access: 1 results) Presentation (8 results) (of which Int'l Joint Research: 6 results)

  • [Journal Article] ATNAS: Automatic Termination for Neural Architecture Search2023

    • Author(s)
      Sakamoto Kotaro、Ishibashi Hideaki、Sato Rei、Shirakawa Shinichi、Akimoto Youhei、Hino Hideitsu
    • Journal Title

      Neural Networks

      Volume: 166 Pages: 446-458

    • DOI

      10.1016/j.neunet.2023.07.011

    • Related Report
      2023 Research-status Report
    • Peer Reviewed / Open Access
  • [Presentation] Viewpoint Representation via Dual-Space Embedding of Object-Tag Data2025

    • Author(s)
      Taisei Sakaguchi, Hideaki Ishibashi, Tetsuo Furukawa
    • Organizer
      The 6th International Symposium on Neuromorphic AI Hardware
    • Related Report
      2024 Annual Research Report
    • Int'l Joint Research
  • [Presentation] On the Visualization of Contextual Meaning in Waka2025

    • Author(s)
      Soichiro Takemura, Hideaki Ishibashi, Tetsuo Furukawa
    • Organizer
      The 6th International Symposium on Neuromorphic AI Hardware
    • Related Report
      2024 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Unsupervised Meta-Modeling of a Set of Dynamical Systems: Latent Variable Estimation in State and Observation Spaces2025

    • Author(s)
      Yuki Tokunaga, Hideaki Ishibashi, Tetsuo Furukawa
    • Organizer
      The 6th International Symposium on Neuromorphic AI Hardware
    • Related Report
      2024 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Visual exploration system for discovering favorite Arita porcelain types2025

    • Author(s)
      Daichi Yoshikane, Ishibashi Hideaki, Tetsuo Furukawa
    • Organizer
      The 6th International Symposium on Neuromorphic AI Hardware
    • Related Report
      2024 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Transfer learning for Gaussian process2025

    • Author(s)
      Shotaro Akaho, Hideaki Ishibashi
    • Organizer
      Further Developments of Information Geometry 2025 (FDIG2025)
    • Related Report
      2024 Annual Research Report
    • Int'l Joint Research
  • [Presentation] レベルセット推定の停止基準2024

    • Author(s)
      石橋 英朗, 松井 孝太, 沓掛 健太朗, 日野 英逸
    • Organizer
      2024年度人工知能学会全国大会(第38回)
    • Related Report
      2024 Annual Research Report
  • [Presentation] モデルやタスクの種類に非依存なマルチタスク学習の理論2023

    • Author(s)
      石橋英朗
    • Organizer
      スマートインフォメディアシステム研究会
    • Related Report
      2023 Research-status Report
  • [Presentation] A stopping criterion for Bayesian optimization by the gap of expected minimum simple regrets2023

    • Author(s)
      Hideaki Ishibashi, Masayuki Karasuyama, Ichiro Takeuchi, Hideitsu Hino
    • Organizer
      The 26th International Conference on Artificial Intelligence and Statistics
    • Related Report
      2022 Research-status Report
    • Int'l Joint Research

URL: 

Published: 2022-04-19   Modified: 2026-01-16  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi