A Deep Look into Trust and MUTUAL Understanding in Multi-Agent Cooperative Game Through XRL
This talk introduces our work on how to build trust and mutual understanding in multi-agent cooperative games through explainable reinforcement learning (XRL). We address two critical layers of trust: human trust in machine intelligence, driven by transparency, and inter-agent trust, grounded in predictability. To improve collaboration and learning efficiency, we introduce Belief-based Predictive Auxiliary Learning (BEPAL), which enables agents to share and refine belief states about the hidden environment. By adding predictive auxiliary tasks, BEPAL enhances belief accuracy, accelerates convergence, and improves performance and stability across diverse multi-agent scenarios such as predator-prey, traffic junction, robotic warehouse, and Google Research Football. Complementing this, VisionMask is a model-agnostic, self-supervised XRL method that generates faithful, contrastive saliency maps to explain visual decision-making without altering the agent’s policy. Together, BEPAL and VisionMask advance the transparency and reliability of multi-agent systems, fostering both human trust and cooperative intelligence. Ongoing work focuses on integrating the two frameworks for interpretable teamwork in complex environments.
Bio: Qinru Qiu received her Ph.D. in electrical engineering from University of Southern California. She is currently a Distinguished Professor in the Department of Electrical Engineering and Computer Science at Syracuse University. Qiu is a fellow of IEEE and a distinguished member of ACM. She also is a recipient of NSF CAREER award in 2009 and IEEE Region 1 Technological Innovation award in 2020. She serves as an associate editor of multiple journals including IEEE Transactions on Neural Networks and Learning Systems, IEEE Transactions on Computers, IEEE Transactions on Cognitive and Developmental Systems, etc. She has also served as a technical program committee member of many conferences including DAC, ICCAD, ISLPED, DATE, etc. She is the director of the NSF Industry University Collaborative Research Center on Alternative Sustainable and Intelligent Computing at Syracuse University.