Research map / Learning theory and algorithms / Combinatorial and search optimization
Searching Large Neighborhoods for Integer Linear Programs with Contrastive Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace imitation loss (positive samples only) with contrastive loss using negative samples (ablation) | LNS policy (GCN, partial features) · MVC-S primal gap | primal gap %: 0.29 -> 0.17 | Helped |
| replace imitation loss with contrastive loss (ablation) | LNS policy (GCN, partial features) · CA-S primal gap | primal gap %: 1.09 -> 0.75 | Helped |
| add GAT and full features to imitation learning (ablation) | IL-LNS · CA-S primal gap | primal gap %: 1.09 -> 1.13 | No clear effect |
| add GAT and full features to imitation learning (ablation) | IL-LNS · MVC-S primal gap | primal gap %: 0.29 -> 0.24 | Helped |
| replace GCN with GAT (ablation) | CL-LNS partial features · MVC-S / CA-S primal gap | primal gap %: 0.17 -> 0.16 (MVC-S); 0.75 -> 0.76 (CA-S) | No clear effect |
| replace GCN with GAT (ablation) | CL-LNS partial features · CA-S primal integral | primal integral: 57.9 -> 53.8 | Helped |
| add full feature set (FF) on top of GAT (ablation) | CL-LNS · CA-S primal gap | primal gap %: 0.76 -> 0.65 | Helped |
| add full feature set on top of GAT (ablation) | CL-LNS · MVC-S primal gap | primal gap %: 0.16 -> 0.17 | No clear effect |
| CL-LNS vs best baseline on large CA instances | CL-LNS · CA-L primal gap | primal gap %: 1.61 (LB-Relax) -> 0.09 | Helped |
| CL-LNS vs Random on MIS-L | CL-LNS · MIS-L primal gap | primal gap %: 0.11 -> 0.12 | Hurt |
| CL-LNS vs RL-LNS on SC-L survival rate | CL-LNS · SC-L survival rate | survival rate | Hurt |
Closest papers by meaning
- A General Large Neighborhood Search Framework for Solving Integer Linear Programs (NeurIPS 2020 · similarity 0.91)
- Large Language Model-driven Large Neighborhood Search for Large-Scale MILP Problems (ICML 2025 · similarity 0.89)
- BTBS-LNS: Binarized-Tightening, Branch and Search on Learning LNS Policies for MIP (ICLR 2025 · similarity 0.89)
- Contrastive Predict-and-Search for Mixed Integer Linear Programs (ICML 2024 · similarity 0.88)
- Regularized Large Neighborhood Search (NeurIPS 2026 · similarity 0.87)
- BIPNN: Learning to Solve Binary Integer Programming via Hypergraph Neural Networks (NeurIPS 2025 · similarity 0.85)
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.