Junxian He

Assistant Professor in The Hong Kong University of Science and Technology

Junxian He is an assistant professor in the department of computer science and engineering at the Hong Kong University of Science and Technology. He received his PhD degree from Carnegie Mellon University, Language Technologies Institute. He serves as the area chair of ICLR, ACL, and EMNLP. His recent research focuses on complex reasoning/planning, reinforcement learning, and agentic aspects of large language models.

Talk

Title: Learning to Model Code Execution for General Reasoning of Language Models

Abstract: In this talk, I will present our work CodeI/O, an approach that condenses reasoning patterns from raw code data to enhance the reasoning abilities of large language models across multiple domains. Specifically, we train large language models to either predict the execution output given a code input or predict the input given an output, effectively serving as a world model for code execution. Our goal is for the model to learn the diverse and complex reasoning processes inherent in code execution. We show that by reasoning in natural language about code execution, the model is able to improve its reasoning capabilities across a wide range of domains, including symbolic, logical, and mathematical reasoning. Notably, this data can be synthesized directly from raw code online without requiring human annotations, opening a new path for generating reasoning-intensive datasets to support generalized reasoning learning.