Research map / Label-efficient and robust learning / Self-supervised and contrastive
MaskCO: Masked Generation Drives Effective Representation Learning and Exploiting for Combinatorial Optimization
Lvda Chen, Yang Li, Junchi Yan
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 mask-and-reconstruct correction to decoding (ablation) | MaskCO · TSP-500 (T=320) | optimality gap: 0.0520% -> 0.0020% | Improved |
| more forward passes inside single construction pass instead of correction (ablation) | MaskCO w/o correction · TSP-500 | optimality gap: 0.8656% -> 0.0520% | Improved |
| use Transformer backbone instead of GCN (ablation) | MaskCO · TSP-500 (T=320) | optimality gap: 0.0212% -> 0.0020% | Improved |
| use Transformer backbone instead of GCN (ablation) | MaskCO · TSP-1000 (T=320) | optimality gap: 0.0457% -> 0.0071% | Improved |
| add masked modeling to GCN backbone (MaskCO-GCN vs Fast T2T) (ablation) | GCN solver · TSP-500 | optimality gap: 0.21% -> 0.0212% | Improved |
| add 2-opt between correction steps (ablation) | MaskCO on CVRP-100 (T=640) · CVRP-100 | optimality gap: 0.808% -> 0.232% | Improved |
| add 2-opt between correction steps (ablation) | MaskCO on TSP-1000 (T=320) · TSP-1000 | optimality gap: 0.0110% -> 0.0071% | Improved |
| add capacity relaxation (ablation) | MaskCO on CVRP-500 (T=640) · CVRP-500 | optimality gap: 3.944% -> 0.813% | Improved |
| self-training on pseudo-labels (D0 -> D2) (ablation) | MaskCO trained on 2-opt labels · TSP-500 | optimality gap: 0.362% -> 0.059% | Improved |
| train on noisy 2-opt data rather than clean optimal data (ablation) | MaskCO · TSP-500 | optimality gap: 0.0020% -> 1.2726% | Hurt |
| increase forward passes T (ablation) | MaskCO · CVRP-1000 | optimality gap: 0.798% -> 0.438% | Improved |
| train on RB graphs, test on ER graphs (cross-distribution) | MaskCO MIS · MIS ER-[700-800] | optimality gap: 0.62% -> 0.72% | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ICAM — ICAM: Interpretable Classification via Disentangled Representations and Feature Attribution Mapping (NeurIPS 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Masking meets Supervision: A Strong Learning Alliance (CVPR 2025 · similarity 0.87)
- From Prototypes to General Distributions: An Efficient Curriculum for Masked Image Modeling (CVPR 2025 · similarity 0.87)
- Train for the Worst, Plan for the Best: Understanding Token Ordering in Masked Diffusions (ICML 2025 · similarity 0.87)
- Understanding and Enhancing Mask-Based Pretraining towards Universal Representations (NeurIPS 2025 · similarity 0.87)
- MaskSense: Confronting the Visual Exploration Trap in Masked Image Generation (NeurIPS 2026 · similarity 0.87)
- Masking Teacher and Reinforcing Student for Distilling Vision-Language Models (CVPR 2026 · 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.
