Reasoning over Sets: Toward Consistent LLM Systems

Abstract: Modern large language models (LLMs) increasingly operate not on single inputs, but over sets of information—collections of generated statements, retrieved documents, or intermediate reasoning steps. However, existing frameworks largely treat these elements independently, overlooking the fact that correctness and reliability often emerge at the set level, where interactions such as contradictions, redundancy, or joint relevance become critical. In this talk, I argue that reasoning over sets is a fundamental challenge for building reliable and controllable LLM systems, and present a unified perspective centered on set-level verification, retrieval, and editing. Bio: Jay-Yoon Lee is an Assistant Professor in the Graduate School of Data Science at Seoul National University (SNU). His research focuses on improving the reliability and controllability of AI systems, particularly large language models, by enabling them to reason over sets of information such as generated candidates, retrieved documents, and intermediate reasoning steps. His work spans structured prediction, constraint-based learning, and retrieval-augmented generation, with recent emphasis on set-level verification, retrieval, and editing. Prior to joining SNU, he was a postdoctoral researcher at the University of Massachusetts Amherst, working with Andrew McCallum. He received his Ph.D. in Computer Science from Carnegie Mellon University under the supervision of Jaime Carbonell, and his B.S. in Electrical Engineering from KAIST. He serves as an Area Chair for ACL, EMNLP, NeurIPS, ICLR, and COLM. He is the recipient of the Excellent Mid-career Researcher Program (NRF of Korea, PI) and leads multiple interdisciplinary projects applying AI to scientific domains, including biology, neuroscience, and biomedical applications. URL to LinkedIn or personal website https://leejayyoon.github.io/