Research map / Image generation / Fast diffusion sampling
Unsupervised Diffusion Solver for Combinatorial Optimization via Combinatorial Adjoint Matching
Shengyu Feng, Tarun Suresh, Yiming Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add adjoint propagation to the TSP training target (ablation) | CAM · TSP-500 | optimality gap: 2.83% -> 2.18% | Improved |
| use 2-OPT instead of greedy decoding for the terminal heuristic (ablation) | CAM · TSP-500 | optimality gap: 2.36% -> 2.18% | Improved |
| use CAM instead of SDDS as unsupervised training objective | diffusion CO solver · MIS ER-[9000-11000] | independent set size: 350.63 -> 384.38 | Improved |
| use CAM instead of RLNN | diffusion CO solver · MIS RB-[800-1200] | independent set size: 39.96 -> 41.25 | Improved |
| use CAM instead of DiffUCO | diffusion CO solver · MIS RB-[800-1200] | independent set size: 40.52 -> 41.25 | Improved |
| use CAM instead of SDDS | diffusion CO solver · TSP-1000 | tour length: 23.90 -> 23.70 | Improved |
| use CAM instead of SDDS | diffusion CO solver · CVRP-100 | route length: 16.16 -> 15.98 | Improved |
| use CAM instead of supervised DIFUSCO | diffusion CO solver · TSP-500 | tour length: 16.85 -> 16.91 | Hurt |
| use CAM instead of supervised DIFUSCO | diffusion CO solver · CVRP-100 | route length: 15.97 -> 15.98 | Hurt |
| use CAM instead of SDDS | diffusion CO solver · Max Cut BA-[800-1200] | cut size: 2963.95 -> 2964.20 | No clear effect |
| use CAM instead of the OR solver MQLib | Max Cut solver · Max Cut BA-[800-1200] | cut size: 2961.38 -> 2964.20 | Improved |
| use CAM instead of the OR solver KaMIS | MIS solver · MIS ER-[9000-11000] (transfer) | independent set size: 381.31 -> 384.38 | Improved |
Closest papers by meaning
- Efficient Combinatorial Optimization via Heat Diffusion (NeurIPS 2024 · similarity 0.89)
- A Diffusion Model Framework for Unsupervised Neural Combinatorial Optimization (ICML 2024 · similarity 0.89)
- AdjointDEIS: Efficient Gradients for Diffusion Models (NeurIPS 2024 · similarity 0.88)
- Fast T2T: Optimization Consistency Speeds Up Diffusion-Based Training-to-Testing Solving for Combinatorial Optimization (NeurIPS 2024 · similarity 0.88)
- Efficient Adjoint Matching for Fine-tuning Diffusion Models (NeurIPS 2026 · similarity 0.88)
- Adjoint Matching: Fine-tuning Flow and Diffusion Generative Models with Memoryless Stochastic Optimal Control (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
