Overview
This theme studies mathematical reasoning in large language models, the benchmarks used to evaluate it, and how reasoning can be reused rather than regenerated. Ongoing directions include reasoning reuse as a paradigm for model collaboration and an adaptive hint generator that tailors guidance to the reasoning process.
Motivation
Mathematical reasoning is a demanding testbed for large language models, and current pipelines often regenerate reasoning from scratch for every problem. This theme studies how to evaluate LLM math reasoning rigorously and how to reuse reasoning across problems and models, reducing redundant computation and enabling more effective collaboration between models.
Ongoing Projects
- LLM reasoning reuse. Developing methods that reuse reasoning rather than regenerating it, framed as a new paradigm for model collaboration.
- Adaptive hint generator. Building a hint generator that adapts the guidance it provides to the model’s reasoning process.
Publications
An ICLR 2026 workshop paper, “Towards Reasoning Reuse: A New Paradigm in Model Collaboration” (Third Workshop on Test-Time Updates, Main Track), introduces the reasoning-reuse direction. See the publications list above for details.
Related Publications
-
Towards Reasoning Reuse: A New Paradigm in Model Collaboration
Zhengxi Li, Fuyuan Lyu, Qiyuan Zhang, Ye Yuan, Haolun Wu, Xue Liu
ICLR 2026 Workshop (Third Workshop on Test-Time Updates, Main Track) · 2026
Impact Holders
Impact holders and user communities will be added as the project scope becomes clearer.