Research Institute for Electronic Science, Hokkaido University


LAST UPDATE 2022/07/13

  • 研究者氏名
    Researcher Name

    田畑公次 Koji TABATA
    准教授 Associate Processor
  • 所属
    Professional Affiliation

    Research Institute for Electronic Science, Hokkaido University

    附属社会創造数学研究センター データ数理研究分野
    Research Center of Mathematics for Social Creativity, Molecule & Life Nonlinear Sciences Laboratory
  • 研究キーワード
    Research Keywords


    Machine learning
    Multi-armed bandit
    Pure exploration problem
Research Subject
Development and application of multi-armed bandit algorithm

研究の背景 Background


Multi-armed bandit is a model in which the agent selects one of the K options called as arms in every time step and observes reward from chosen arm. To achieve a given goal, for example, "maximizing cumulative rewards up to a certain time", it is necessary to consider the trade-off between knowledge exploitation and exploration. This model has been applied in various area in recent years, such as online advertising, speeding up and efficiency of measurement, and game AI.

研究の目標 Outcome


I am working on formulation of mathematical problem settings to improve diagnosis efficiency using Raman measurement, development of multi-armed bandit algorithm to realize measurement with guaranteed accuracy, etc. My target is to realize various systems in which measurement and information are highly integrated.

研究図Research Figure

Fig.1. finding a good candidate by multi armed bandit armed bandit.

Fig.2. Accelerated diagnosis by intensively measuring areas with high disease indicators.

文献 / Publications

Koji Tabata, Atsuyoshi Nakamura, Junya Honda and Tamiki Komatsuzaki, “A bad arm existence checking problem: How to utilize asymmetric problem structure?”, Machine Learning, 109(2), 327-372, 2020.