Yali Du (King's College London)

Title: Reinforcement Learning with Human Values

Abstract: In the coming years, diverse ecologies of AI systems are envisioned to rapidly and complexly interact with each other and with humans. Collaborative industrial robots will work on factory floors alongside laborers, care robots will assist human health workers, and personal AI assistants will help with scheduling, albeit in an elementary way. Therefore, it is essential to develop AI systems that can effectively and reliably collaborate with humans in various contexts. This gives rise to the agent alignment problem: how do we create agents that behave in accordance with the user's intentions and possess the ability to adapt to changing circumstances? In this talk, I will provide an overview of reinforcement learning and its challenges in reliable reward design, human coordination and safety guarantees. I will then discuss my attempts in agent alignment to address these issues by improving data efficiency for human labels in preference-based reinforcement learning, and enhancing safety of the agent learning process through human feedback for control and games.

Bio: Yali Du is a Lecturer (Assistant Professor) at King's College London. Her research aims to enable machines to exhibit cooperative and trustworthy behavior in intelligent decision-making tasks. Her work focuses on reinforcement learning and multi-agent cooperation, with interests in topics such as generalization, zero-shot coordination, evaluation of human and AI players, and social agency (e.g., human-involved learning, safety, and ethics). For her work in reinforcement learning and cooperation, she was chosen for the AAAI New Faculty Highlights program (2023), Rising Star in AI (KAUST 2023). Prior to her current position, Yali received her PhD from the University of Technology Sydney and worked as a postdoc at University College London.