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

Machine learning problems as retrieval in high dimensional space

Research Project

Project/Area Number 19H04173
Research Category

Grant-in-Aid for Scientific Research (B)

Allocation TypeSingle-year Grants
Section一般
Review Section Basic Section 61030:Intelligent informatics-related
Research InstitutionChiba Institute of Technology

Principal Investigator

SHIMBO Masashi  千葉工業大学, 人工知能・ソフトウェア技術研究センター, 主席研究員 (90311589)

Co-Investigator(Kenkyū-buntansha) 重藤 優太郎  千葉工業大学, 人工知能・ソフトウェア技術研究センター, 主任研究員 (50803392)
Project Period (FY) 2019-04-01 – 2023-03-31
Project Status Completed (Fiscal Year 2022)
Budget Amount *help
¥17,030,000 (Direct Cost: ¥13,100,000、Indirect Cost: ¥3,930,000)
Fiscal Year 2022: ¥4,160,000 (Direct Cost: ¥3,200,000、Indirect Cost: ¥960,000)
Fiscal Year 2021: ¥4,160,000 (Direct Cost: ¥3,200,000、Indirect Cost: ¥960,000)
Fiscal Year 2020: ¥4,160,000 (Direct Cost: ¥3,200,000、Indirect Cost: ¥960,000)
Fiscal Year 2019: ¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000)
Keywords近傍検索 / 知識グラフ / 表現学習 / 事前学習 / 非相関化 / 埋め込み / 自己教師あり学習 / 正則化 / 高次元空間 / 知識グラフ補完 / 自己教師付き学習 / 機械学習 / データマイニング
Outline of Research at the Start

本研究では, 機械学習のさまざまなタスクを, 高次元空間における検索の問題として再解釈する. 具体的には, 最近傍検索を行う際に障害となる, 高次元空間特有のバイアス (「空間中心性」) の存在に着目する. そのうえで, 上記機械学習タスクに対してこれまで提案された各種アプローチを, このバイアスを軽減する距離・類似度尺度の設計という観点から再評価する. 同時に, その悪影響を直接軽減する手法を開発し, タスク性能の向上につなげる.

Outline of Final Research Achievements

The main results in this project are as follows:
(1) Improvement of knowledge graph embedding. Specifically, we pointed out that the existing models for knowledge graph embedding are unsuitable for queries called "path queries," and proposed a model that overcome this shortcoming. Also, to reduce memory usage during inference, we proposed a binary quantization method for knowledge graph embedding.
(2) We pointed out that the recent non-contrastive learning models for image representation are inefficient for high-dimensional embeddings, and proposed a scalable alternative that removes this drawback while maintaining downstream accuracy.

Academic Significance and Societal Importance of the Research Achievements

多くのニューラルネットの応用分野において, 外部知識を, 知識グラフとして表現して活用する手法が多く提案されており, 我々の成果はこういったアプローチの補助となる. 表現学習に関する提案法は, スケーラブルな特徴量非相関化法であるが, これは画像のみならず多様なデータの表現学習にもほぼそのまま適用可能である. また, 提案法は特徴量の非相関化が必要な表現学習以外のタスクにも広く用いることができる.

Report

(5 results)
  • 2022 Annual Research Report   Final Research Report ( PDF )
  • 2021 Annual Research Report
  • 2020 Annual Research Report
  • 2019 Annual Research Report
  • Research Products

    (9 results)

All 2023 2021 2020 2019 Other

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

  • [Journal Article] Binarized Embeddings for Fast, Space-Efficient Knowledge Graph Completion2021

    • Author(s)
      Katsuhiko Hayashi、Koki Kishimoto、Masashi Shimbo
    • Journal Title

      IEEE Transactions on Knowledge and Data Engineering

      Volume: -- Pages: 1-13

    • DOI

      10.1109/tkde.2021.3075070

    • Related Report
      2021 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Transductive Data Augmentation with Relational Path Rule Mining for Knowledge Graph Embedding2021

    • Author(s)
      Hirose Yushi、Shimbo Masashi、Watanabe Taro
    • Journal Title

      Proceedings of the 12th IEEE International Conference on Big Knowledge

      Volume: なし Pages: 377-384

    • DOI

      10.1109/ickg52313.2021.00057

    • Related Report
      2021 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] A Greedy Bit-flip Training Algorithm for Binarized Knowledge Graph Embeddings2020

    • Author(s)
      Hayashi Katsuhiko、Kishimoto Koki、Shimbo Masashi
    • Journal Title

      Findings of the Association for Computational Linguistics: EMNLP 2020

      Volume: - Pages: 109-114

    • DOI

      10.18653/v1/2020.findings-emnlp.10

    • Related Report
      2020 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Distant supervision for relation extraction via piecewise attention and bag-level contextual inference2019

    • Author(s)
      Van-Thuy Phi, Joan Santoso, Van-Hien Tran, Hiroyuki Shindo, Masashi Shimbo, and Yuji Matsumoto
    • Journal Title

      IEEE Access

      Volume: 7 Pages: 103570-103582

    • DOI

      10.1109/access.2019.2932041

    • Related Report
      2019 Annual Research Report
    • Open Access / Int'l Joint Research
  • [Presentation] Learning Decorrelated Representations Efficiently Using Fast Fourier Transform2023

    • Author(s)
      Y. Shigeto / M. Shimbo / Y. Yoshikawa / A. Takeuchi
    • Organizer
      CVPR 2023 (To appear)
    • Related Report
      2022 Annual Research Report
    • Int'l Joint Research
  • [Presentation] Binarized knowledge graph embeddings2019

    • Author(s)
      Koki Kishimoto, Katsuhiko Hayashi, Genki Akai, Masashi Shimbo, and Kazunori Komatani
    • Organizer
      41st European Conference on Information Retrieval (ECIR ’19)
    • Related Report
      2019 Annual Research Report
    • Int'l Joint Research
  • [Presentation] A non-commutative bilinear model for answering path queries in knowledge graphs2019

    • Author(s)
      Katsuhiko Hayashi and Masashi Shimbo
    • Organizer
      Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP '19)
    • Related Report
      2019 Annual Research Report
    • Int'l Joint Research
  • [Presentation] A Simple Reranking Method for Knowledge Graph Completion2019

    • Author(s)
      Lu Yuxun, Yutaro Shigeto, Katsuhiko Hayashi, Masashi Shimbo
    • Organizer
      情報処理学会 第241回自然言語処理研究会
    • Related Report
      2019 Annual Research Report
  • [Remarks]

    • URL

      https://stair.center/archives/research/learning-decorrelated-representations-efficiently-using-fast-fourier-transform-cvpr-2023

    • Related Report
      2022 Annual Research Report

URL: 

Published: 2019-04-18   Modified: 2024-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi