知の物理学研究センター / Institute for Physics of Intelligence (iπ)

【日時/Date】
７月１９日（金）10:30-12:00　オンライン開催

【発表者/Speaker】
本武 陽一 氏（一橋大学）

【タイトル/Title】
"Interpretable AI to assist scientists' insight into large pattern dynamics"

【概要/Abstract】
To understand and predict the complex systems with large degrees of freedom observed in nature, it is useful to construct a reduced model of such systems. For example, in a historical success story, the molecular motion of a gas, a large-degree-of-freedom system, is expressed in a reduced coordinate system consisting of its statistics, and a thermodynamic model, a reduced model, has been constructed to understand and predict the phenomenon. Usually, reduced modeling has been based on the insights of scientists. On the other hand, the complex pattern dynamics found in materials such as magnetic and polymeric materials, electromagnetic turbulence, and active matter sometimes make reduced modeling that relies solely on such insights difficult. In this talk, we will introduce our research to realize understanding and prediction of such phenomena by reduced modeling of complex pattern dynamics using machine learning methods such as deep learning [1, 2, 3]. Furthermore, through those studies, we explain how the nature of many physics models of complex phenomena as singular models in statistical science makes them an obstacle to interpretable physics modeling by machine learning models, and we also present research on the development of a physics model evaluation framework to overcome the nature of such physics models[4].

[1] Y. Mototake, Phys. Rev. E 103, 033303 (2021).
[2] S.Tsuji, R. Murakami, H. Shouno, Y. Mototake, NeurIPS 2023 Workshop on Machine Learning and Physical Science (2023).
[3] Y. Mototake, M. Mizumaki, K. Kudo, K. Fukumizu, arXiv:2204.12194 (2022).
[4] K. Nagata, Y. Mototake, arXiv:2406.00369 (2024).

これらの講演に関するZoomのリンク等の案内を受け取ることを希望されるかたは、下記のgoogle formからメールアドレスをご記入ください。こちらに登録頂いた情報は、案内の配信のみに利用いたします。

登録フォーム：https://forms.gle/xnLmd9Kc1BaaNPgq8

過去の発表リスト：https://www.phys.s.u-tokyo.ac.jp/about/17106/

　　　　　　　　　　世話人：知の物理学研究センター　髙橋昂

⇦Top page