Date and Time: April 27, 2026, 10:30 -- 11:30 (JST)
Venue: Hybrid
*Open Space is available to AIP researchers only
Title: Building AI Agents That Act, Learn, and Evolve
Speaker:
Prof. Huaxiu Yao (University of North Carolina at Chapel Hill)
Abstract:
Current large language models are static after deployment: they cannot practice in realistic environments, accumulate skills from experience, or improve over time. This talk presents a full-stack approach to building AI agents that act, learn, and evolve. We introduce Agent World Model, which automatically synthesizes thousands of executable training environments from seed names alone, enabling large-scale agent reinforcement learning. Within these environments, SkillRL trains agents to extract reusable skills from both successes and failures. MetaClaw extends this evolution beyond training into deployment, enabling agents to learn from natural interactions and fine-tune during idle time with zero downtime. To support long-running agents, SimpleMem provides lifelong memory through semantic compression, significantly reducing token cost while achieving state-of-the-art performance across text, image, audio, and video. ClawArena complements these systems with a benchmark for evolving information environments, where sources contradict and evidence changes, revealing that adaptability remains a critical gap even in frontier models. Together, these components form a blueprint for AI agents that practice, remember, evolve, and are evaluated under realistic conditions.
Short Bio:
Huaxiu Yao is a tenure-track Assistant Professor in the Department of Computer Science at the University of North Carolina at Chapel Hill, with a joint appointment in the School of Data Science and Society. He was previously a Postdoctoral Scholar in Computer Science at Stanford University, where he worked with Chelsea Finn. His current research focuses on developing agentic multimodal foundation models that are widely generalizable and well-aligned with human preferences. He is also committed to applying these methods to real-world data science challenges in domains such as robotics and biomedicine. Dr. Yao has authored over 80 publications in top machine learning venues, including ICML, ICLR, and NeurIPS, and has served as a (Senior) Area Chair and workshop organizer at conferences such as ICML, NeurIPS, ICLR, ACL, and EMNLP. His research has been recognized with several honors, including Outstanding Paper Award in TMLR, KDD 2024, ICLR 2026 RSI Workshop, Amazon Research Award, Cisco Faculty Award, AAAI 2024 New Faculty Highlights, PharmAlliance Early Career Researcher Award, NC TraCS Innovation to Impact Awards, and R.J. Reynolds Junior Faculty Development Award.
