Professor Jingrui He's lab received an outstanding paper award at the Multi-Modal Reasoning for Agentic Intelligence Workshop, which was held during the International Conference on Computer Vision (ICCV 2025) last month in Honolulu, Hawaii.
The paper, "AutoTool: Dynamic Tool Selection and Integration for Agentic Reasoning," was co-authored by University of Illinois researchers Jiaru Zou, MS student in computer science; Sirui Chen, PhD student in computer science; information sciences PhD students Yunzhe Qi and Mengting Ai; and Professor Jingrui He, and Princeton University researchers Ling Yang, Ke Shen, and Mengdi Wang.
"We research in the area of Agentic AI and built a framework that teaches large language models to dynamically select and integrate tools as they reason, instead of relying on a fixed set," said lead author Zou. "From solving math problems to writing code and interpreting images, AutoTool marks a step toward more adaptive and self-directed AI systems."
In their paper, the researchers demonstrate how AutoTool consistently outperforms advanced large language model agents and tool-integration methods in math and science reasoning, search-based QA, code generation, and multimodal understanding. The new framework also exhibits stronger generalization by using new tools as they become available during operation rather than being limited to a fixed set of tools it was trained with.