• 研究課題をさがす
  • 研究者をさがす
  • KAKENの使い方
  1. 課題ページに戻る

2020 年度 実施状況報告書

Inverse materials design by integrating transfer learning techniques into a Bayesian framework

研究課題

研究課題/領域番号 18K18017
研究機関統計数理研究所

研究代表者

Wu Stephen  統計数理研究所, データ科学研究系, 准教授 (70804186)

研究期間 (年度) 2018-04-01 – 2022-03-31
キーワードTransfer learning / Materials informatics / Polymer design / Open source software
研究実績の概要

In 2020, I have started the second round of design process for high thermal conductivity polymers. I focused more on actual industrial application, targeting a wider range of material properties based on the manufacturing needs, such as linear thermal expansion coefficient, dielectric constant, dielectric loss tangent, water absorption, etc. Meanwhile, I narrowed down the search of polymers to a few classes of polymers, such as polyimide and liquid crystal polymers. New predictive models that take into account of descriptors built from physical and chemical properties were developed to enhance the predictive power of models using purely fingerprints of molecules. As a result, new candidates were identified and now under-going synthetic path evaluation by my collaborators. Furthermore, the transfer learning technique developed in this study has been modified and applied to two other studies targeting microscopic images of crystal materials and concrete crack segmentation. Two peer-reviewed journal papers were published as a result. Although international travel was prohibited due to COVID19, I have participated in two online talks related to the extension projects of transfer learning.

現在までの達成度 (区分)
現在までの達成度 (区分)

3: やや遅れている

理由

While the computational design of the second round of high thermal conductivity polymers has been progressing smoothly, the synthesis of new polymers has been delayed due to the COVID19 hazard. I was not able to visit my collaborators who can synthesize the polymers, therefore, the synthesis progress is delayed. Furthermore, the COVID19 situation also limited the supply of ingredients for the synthesis tasks. As a result, I used the extra time to modify and apply my transfer learning technique to two other problems targeting microscopic images of crystal materials and concrete crack segmentation.

今後の研究の推進方策

In the coming fiscal year, the top priority is to push the progress of polymer synthesis. Bayesian optimization techniques can be applied to filter out a small subset of polymer candidates that are considered to be at the highest priority for synthesis. Reliable remote communication platform has been secured to ensure smooth communication with my collaborators in order to progress the synthesis plans. Meanwhile, I will begin building a larger publicly available data and model library of different classes of polymers to facilitate the use of data science methods in the industry of materials science. The final goal is to cover a wide range of industrial applications through inclusion of more material properties, such as dielectric constant or refractive index.

次年度使用額が生じた理由

Due to COVID19, traveling is prohibition in 2020. In-person collaboration and presentation may be resumed in the next fiscal year, which will be the main use of the fund extended to the next year. Also, upon success of newly synthesized polymers, journal papers will be published which may require the use of the remaining funds.

  • 研究成果

    (7件)

すべて 2021 2020

すべて 雑誌論文 (5件) (うち国際共著 5件、 査読あり 4件、 オープンアクセス 3件) 学会発表 (2件) (うち招待講演 1件)

  • [雑誌論文] Recovering compressed images for automatic crack segmentation using generative models2021

    • 著者名/発表者名
      Huang Yong、Zhang Haoyu、Li Hui、Wu Stephen
    • 雑誌名

      Mechanical Systems and Signal Processing

      巻: 146 ページ: 107061~107061

    • DOI

      10.1016/j.ymssp.2020.107061

    • 査読あり / 国際共著
  • [雑誌論文] Data augmentation in microscopic images for material data mining2020

    • 著者名/発表者名
      Ma Boyuan、Wei Xiaoyan、Liu Chuni、Ban Xiaojuan、Huang Haiyou、Wang Hao、Xue Weihua、Wu Stephen、Gao Mingfei、Shen Qing、Mukeshimana Michele、Abuassba Adnan Omer、Shen Haokai、Su Yanjing
    • 雑誌名

      npj Computational Materials

      巻: 6 ページ: 125

    • DOI

      10.1038/s41524-020-00392-6

    • 査読あり / オープンアクセス / 国際共著
  • [雑誌論文] Potentials and challenges of polymer informatics: exploiting machine learning for polymer design2020

    • 著者名/発表者名
      Stephen Wu, Hironao Yamada, Yoshihiro Hayashi, Massimiliano Zamengo, Ryo Yoshida
    • 雑誌名

      arXiv (in press at Proceedings of the Institute of Statistical Mathematics (2021 special issue))

      巻: NA ページ: NA

    • オープンアクセス / 国際共著
  • [雑誌論文] A general class of transfer learning regression without implementation cost2020

    • 著者名/発表者名
      Shunya Minami, Song Liu, Stephen Wu, Kenji Fukumizu, Ryo Yoshida
    • 雑誌名

      Proceedings of the AAAI Conference on Artificial Intelligence

      巻: NA ページ: NA

    • 査読あり / 国際共著
  • [雑誌論文] Bayesian Algorithm for Retrosynthesis2020

    • 著者名/発表者名
      Guo Zhongliang、Wu Stephen、Ohno Mitsuru、Yoshida Ryo
    • 雑誌名

      Journal of Chemical Information and Modeling

      巻: 60 ページ: 4474~4486

    • DOI

      10.1021/acs.jcim.0c00320

    • 査読あり / オープンアクセス / 国際共著
  • [学会発表] Recovering compressed images for auto-segmentation of building cracks using deep generative models2020

    • 著者名/発表者名
      Stephen Wu
    • 学会等名
      The 163rd TCU-ARL Seminar: International Workshop on Data-driven Infrastructure Maintenance and Risk Management
  • [学会発表] Recovering compressed images for auto-segmentation of building cracks using deep generative models2020

    • 著者名/発表者名
      Stephen Wu
    • 学会等名
      応用力学講演会 2020
    • 招待講演

URL: 

公開日: 2021-12-27  

サービス概要 検索マニュアル よくある質問 お知らせ 利用規程 科研費による研究の帰属

Powered by NII kakenhi