Research map / Label-efficient and robust learning / Recognition and long-tail
Dynamic Stratified Contrastive Learning with Upstream Augmentation for MILP Branching
Tongkai Lu, Shuai Ma, Chongyang Tao
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 upstream-augmented MILP derivation (ablation) | SC-MILP · Set Covering | acc@1 (%): 64.2 -> 68.9 | Improved |
| add equivalent MILP derivation (linear transformation, redundant constraint) (ablation) | SC-MILP · Set Covering | acc@1 (%): 64.8 -> 68.9 | Improved |
| add perturbed MILP derivation (ablation) | SC-MILP · Set Covering | acc@1 (%): 68.1 -> 68.9 | Improved |
| add dynamic stratified contrastive learning (ablation) | SC-MILP · Set Covering Hard | solving time (s): 1137.06 -> 953.24 | Improved |
| add dynamic stratified weights to contrastive loss (ablation) | SC-MILP · Set Covering | acc@1 (%): 66.4 -> 68.9 | Improved |
| add hybrid branching strategy (top-k then strong branching at upstream nodes) (ablation) | SC-MILP · Set Covering Hard | solving time (s): 1011.35 -> 953.24 | Improved |
| add hybrid branching strategy (ablation) | SC-MILP · Set Covering Easy | solving time (s): 5.95 -> 5.99 | No clear effect |
| train GCNN on upstream-augmented data instead of equal-sized original data (ablation) | GCNN · Set Covering | acc@1 (%): 62.1 -> 65.4 | Improved |
| train MILP-Evolve on upstream-augmented data instead of equal-sized original data (ablation) | MILP-Evolve · Set Covering | acc@1 (%): 61.7 -> 63.6 | Improved |
| add dynamic stratified contrastive learning on top of upstream-augmented data (ablation) | GCNN · Set Covering Hard | solving time (s): 1154.33 -> 953.24 | Improved |
| increase number of strata m (ablation) | SC-MILP · four generated benchmarks | solving time | Mixed |
| increase hybrid ratio rho (ablation) | SC-MILP · four generated benchmarks | accuracy and time | Mixed |
Closest papers by meaning
- CAMBranch: Contrastive Learning with Augmented MILPs for Branching (ICLR 2024 · similarity 0.92)
- RoME: Domain-Robust Mixture-of-Experts for MILP Solution Prediction across Domains (NeurIPS 2025 · similarity 0.86)
- Rethinking the Capacity of Graph Neural Networks for Branching Strategy (NeurIPS 2024 · similarity 0.86)
- Towards Foundation Models for Mixed Integer Linear Programming (ICLR 2025 · similarity 0.85)
- Contrastive Predict-and-Search for Mixed Integer Linear Programs (ICML 2024 · similarity 0.84)
- MILP-FBGen: LP/MILP Instance Generation with Feasibility/Boundedness (ICML 2024 · similarity 0.84)
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.
