Reproduce “UCPO: A Universal Constrained Combinatorial Optimization Method via Preference Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhanhong Fang, Debing Wang, Jinbiao Chen, Jiahai Wang, Zizhen Zhang
UCPO fine-tunes pretrained neural solvers for constrained routing by ranking sampled solutions: feasible routes outrank infeasible ones, feasible routes are ordered by objective, and infeasible routes by a Lagrangian score. It adds this preference-based loss without changing architecture. Across four benchmark problems at 50 and 100 nodes, it improves feasibility and route quality, but does not guarantee feasible outputs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.