LLM Reasoning in Combinatorial Optimization
Investigating the boundaries of LLM reasoning for NP-hard problems
LLM Reasoning in Combinatorial Optimization
Status: Ongoing Research
Affiliation: THU-C3I, Tsinghua University
Duration: October 2025 - Present
Overview
This research investigates the boundaries of Large Language Model (LLM) reasoning in combinatorial optimization problems, with the goal of enabling LLMs to tackle NP-hard problems while ensuring robust generalization across diverse domains.
Research Goals
- Explore the capabilities and limitations of LLM reasoning in solving combinatorial optimization problems
- Develop methods to enable LLMs to handle NP-hard computational problems
- Ensure robust generalization across different problem domains
- Bridge the gap between natural language understanding and formal problem-solving
Research Context
This work builds upon my background in competitive programming and problem-solving, applying structured algorithmic thinking to advance LLM reasoning capabilities. The research aligns with my interest in building general intelligent systems capable of accelerating scientific discovery through verifiable reasoning.
Supervision
Supervised by Prof. Ning Ding and Dr. Youbang Sun at THU-C3I, Tsinghua University.
This research represents the intersection of my interests in LLM Reasoning, structured problem-solving, and building general intelligent systems with verifiable reasoning capabilities.