Runyu Zhang
Harvard University
runyuzhang@fas.harvard.edu
Bio
Runyu (Cathy) Zhang is a Ph.D. candidate (2019 – present) majoring in applied mathematics in the School of Engineering and Applied Sciences, Harvard University. During the summer of 2022, she served as a research intern at Salesforce Research, working on multi-agent reinforcement learning. Prior to pursuing her Ph.D., she earned a B.S. degree in mathematics from Peking University in 2019. Her research interests lie broadly in online control methods, reinforcement learning, game theory and optimization, with particular focus on multi-agent systems. She has received the Certificate of Distinction and Excellence in Teaching at Harvard University and was a finalist of the Two Sigma Diversity PhD Fellowship in 2022.
Areas of Research
- Machine Learning
Efficient and Resilient Coordination of Multi-agent Systems
My research centers round distributed control, multi-agent reinforcement learning (RL), and risk-sensitive/safe decision-making, with a focus on developing robust and practical algorithms for complex systems. Distributed control is essential for managing large-scale, interconnected systems where centralized control is impractical or impossible. My work in this area aims to create scalable algorithms that enable efficient coordination among distributed agents, ensuring stability and performance even in the presence of communication delays and uncertainties. In the scenario of multi-agent RL, I investigate the unique challenges that arise when multiple agents interact in a shared environment. These include non-stationarity due to simultaneous learning, coordination difficulties, and the complexity of finding and maintaining Nash equilibria. I have designed efficient NE-seeking algorithms and analyzed their convergence rates in various game-theoretic settings, including Markov potential games and general-sum stochastic games. Furthermore, I have introduced the concept of smooth stochastic games (SGs) to establish optimality guarantees, such as bounding the price of anarchy for equilibrium outcomes. The third topic of my research is risk-sensitive and safe decision-making, crucial for tasks like online decision making and the control of dynamical systems under uncertainty. I am actively exploring the development of sample-efficient algorithms that incorporate robustness, risk-sensitivity, and safety, ensuring reliable performance even in the face of model uncertainties or estimation errors. By integrating these principles into both distributed control and multi-agent RL, my research aims to design algorithms for autonomous systems that can operate safely and efficiently in uncertain, dynamic and distributed environments.