CIIR Talk Series: Huan Sun, Smarter and Safer Every Day? Continual Learning and Safety in Computer-Use Agents

Content

Image

Speaker

Huan Sun, Ohio State University

Abstract

Computer-use agents are increasingly deployed in open-ended digital environments — interacting with the web and executing multi-step tasks across applications. Yet real-world deployment exposes a fundamental tension: agents must continually adapt to unseen scenarios and distribution shifts, while their safety and security grow more critical as their capabilities increase. In this talk, I will discuss our recent efforts toward continual learning in computer-use agents for environment adaptation, proactive discovery of long-tail unintended behaviors without adversarial attacks, and more fundamentally, understanding transformers’ limitations and potential for OOD generalization in multi-hop problem solving. Together, these efforts help build agents that are both capable and safe in deployment.

Speaker Bio

Huan Sun is an associate professor in the Department of Computer Science and Engineering and an endowed College of Engineering Innovation Scholar at The Ohio State University. Her research focuses on advancing both the capability and safety of LLM-based agents, with an emphasis on web agents, computer-use agents, and agents for data-driven scientific discovery. Huan has led or co-led a series of foundational projects in this space with broad adoption in academia and industry.  She is a recipient of the NSF CAREER Award and has received multiple paper awards from leading venues, including ACL, SIGMOD, CVPR.