Guest Talk: Modelling collective behavior with deep reinforcement learning by Kazushi Tsutsui

Monday, 18. September 2023
15:30 - 16:30



Kazushi Tsutsui

Collective behavior of animals in nature, such as cooperation and competition, is a complex phenomenon that involves a nonlinear dynamical process. Therefore, it is often difficult to verify how observed group behaviors are formed and whether the formed behaviors benefit individuals in the group. In this talk, I will discuss how we use deep reinforcement learning to model cooperative hunting for a group of chimpanzees. Using the proposed method, we showed that collaborative hunting, which has been thought to require a high-level cognition, can emerge with a simple decision process of mapping between observations and actions via distance-dependent internal representations. This suggests that collaborative hunting does not necessarily rely on complex cognitive processes, such as anticipating others’ movement or sharing intentions among predators. Our computational approach has the potential to complement findings from observations in nature and could lead to a comprehensive understanding of complex phenomena.

Kazushi Tsutsui is a designated assistant professor at the Institute for Advanced Research / Graduate School of Informatics, Nagoya University, Japan. He worked on machine learning research at Nagoya University in 2019 before joining the faculty in 2020. He is currently leading young researcher units at Nagoya University for the advancement of new fields. His work involves modeling multi-agent behavior in nature/sports to understand computational mechanisms of collective animal and human behavior. Recently, he has also studied the neuronal basis for collective behavior through collaboration with neuroscientists on Drosophila and with brain scientists on team play in sports.