Pre-Conference Talk by BUI The Viet | Imitation Learning in Cooperative Multiagent Games

Imitation learning (IL) is key to boosting agent performance in complex multiagent environments. In cooperative games, the challenges of high-dimensional spaces and inter-agent dependencies require advanced strategies. This talk discusses integrating IL techniques from "Mimicking To Dominate" and "Inverse Factorized Soft Q-Learning." The first approach enhances multi-agent reinforcement learning (MARL) by predicting opponent actions with hidden actions and local observations, achieving superior results in environments like SMACv2. The second extends inverse soft-Q learning to multi-agent settings using centralized learning with mixing networks, ensuring efficient learning. By combining these methods, we propose a framework that improves prediction and efficiency, outperforming current algorithms and advancing adaptive multiagent systems. 

This is a Pre-Conference talk for The Thirty-Eighth Annual Conference on Neural Information Processing Systems (NeurIPS 2024).

BUI The Viet is a first-year PhD student at the School of Computing and Information Systems, Singapore Management University, under the supervision of MAI Anh Tien. He is a recipient of the Presidential Doctoral Fellowship at SMU. His primary research areas include imitation learning and multi-agent reinforcement learning.