[2C6-OS-7c] OS-7 (3)

[2C6-OS-7c-02] Data-driven modeling in human collective motions

〇Keisuke Fujii^1,2 (1. Nagoya University, 2. RIKEN)

Keywords:Machine Learning, Multi-agent, Dynamical Systems

Modeling and understanding collective motions in which elements complexly interact is an important problem in engineering, physics, and biology. However, in real-world organisms, the elements are not physically connected to each other, and the rules behind them are often unknown. Therefore, data-driven approaches of estimating and understanding the mechanism of collective motions are effective. Here, I will introduce various approaches to solve this problem, and as an example, introduce a graph dynamic mode decomposition that extracts dynamical property in a dynamic network of multi-agent interactions. In the experiment, we classified the team offense and defense strategies with higher accuracy than the existing methods, and clarified the mode of label-dependent individual interactions. In the presentation, I would like to introduce other approaches that are currently being taken.