This work develops an LLM-based optimization framework ensuring strict constraint satisfaction in network optimization. While LLMs possess contextual reasoning capabilities, existing approaches often fail to enforce constraints, causing infeasible solutions. Unlike conventional methods that address average constraints, the proposed framework integrates a natural language-based input encoding strategy to restrict the solution space and guarantee feasibility. For multi-access edge computing networks, task allocation is optimized while minimizing worst-case latency. Numerical evaluations demonstrate LLMs as a promising tool for constraint-aware network optimization, offering insights into their inference capabilities.
@article{arxiv.2509.07492,
title = {Constraint-Compliant Network Optimization through Large Language Models},
author = {Youngjin Song and Wookjin Lee and Hong Ki Kim and Sang Hyun Lee},
journal= {arXiv preprint arXiv:2509.07492},
year = {2025}
}