Events Detail

Speaker:
Keisuke Fujii

Understanding real-world multi-agent motions is a fundamental issue in engineering, biology, and human behavioral science. However, in real-world multi-agent systems, the rules behind their complex movements are often unknown. In such cases, it can be effective to estimate the unknown modules using machine learning from measured data. In this talk, first, I will introduce a machine learning method for estimating interaction rules from movement trajectories in multiple species of animals and a theoretical model of animal behaviors. Next, I will introduce a causal inference method for estimating time-varying individual treatment effects for multi-agent trajectories in different domains, such as biological movement simulators, autonomous driving simulators, and team sports. Lastly, I will introduce a multi-agent reinforcement learning from real-world demonstrations and computer vision problems in sports with a few cameras. These frameworks would provide machine learning-based analysis to discover new insights or multi-agent rules from real-world trajectory data with fewer scientific experiments.

Keisuke Fujii is an associate professor at Nagoya University. He also serves as a visiting scientist at the RIKEN Center for Advanced Intelligence Project in Japan. His research interests include machine learning, multi-agent systems, sports sciences, mathematical models, behavioral sciences, computational biology, and robotics. In particular, he is interested in the integration of domain knowledge and machine learning for analyzing multi-body time series data, such as various sports, animal group behaviors, and autonomous vehicle simulations.