Research map / Learning theory and algorithms / Combinatorial and search optimization
Learning to Search in Local Branching
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 |
|---|---|---|---|
| Use scaled-regression initialization and RL neighborhood adaptation instead of default local branching | Local branching on SC, MIS, and CA · SC, MIS, CA (first and root starts) | primal integral: 56.641 -> 2.874; 4.986 -> 1.125; 7.523 -> 2.780 | Helped |
| Use scaled-regression initialization and RL neighborhood adaptation instead of default local branching | Local branching on SC, MIS, and CA · SC, MIS, CA (first and root starts) | final primal gap (percentage): 1136.74 -> 1.15; 0.20 -> 0.23; 2.64 -> 0.40 | Mixed |
| Use scaled-regression initialization and RL neighborhood adaptation instead of default local branching | Local branching on larger instances · LSC, LMIS, LCA | primal integral: 59.306 -> 2.424; 23.626 -> 1.653; 27.598 -> 9.047 | Helped |
| Use scaled-regression initialization and RL neighborhood adaptation instead of default local branching | Local branching on GISP · GISP (first and root starts) | primal integral: 22.176 -> 13.786; 18.677 -> 9.591 | Helped |
| Use scaled-regression initialization and RL neighborhood adaptation instead of default local branching | Local branching on MIPLIB · MIPLIB (first and root starts) | primal integral: 14.984 -> 11.655; 7.363 -> 5.408 | Helped |
| Use regression-predicted neighborhood size for the first iteration (ablation) | Local branching on SC, MIS, and CA · SC, MIS, CA (first and root starts) | primal integral: 56.641 -> 3.282; 4.986 -> 1.575; 7.523 -> 5.400 | Helped |
| Train the regression model on a mixed SC, MIS, and CA dataset rather than a homogeneous dataset (ablation) | Scaled-regression local branching · SC, MIS, CA | primal integral: 3.282 -> 3.910; 1.575 -> 1.473; 5.400 -> 5.427 | Mixed |
| Add RL neighborhood adaptation to mixed-data regression (ablation) | Scaled-regression local branching · SC, MIS, CA | primal integral: 3.910 -> 2.874; 1.473 -> 1.125; 5.427 -> 2.780 | Helped |
| Use RL neighborhood adaptation with a default initial size, without learned regression initialization (ablation) | Local branching · SC, MIS, CA | final primal gap (percentage): 1136.74 -> 7.05; 0.28 -> 0.37; 2.64 -> 0.49 | Mixed |
| Use the hybrid method instead of regression alone (ablation) | Local branching with mixed-data regression · SC, MIS, CA | final primal gap (percentage): 1.44 -> 1.15; 0.25 -> 0.25; 2.07 -> 0.40 | Mixed |
Closest papers by meaning
- Large Language Model-driven Large Neighborhood Search for Large-Scale MILP Problems (ICML 2025 · similarity 0.88)
- CAMBranch: Contrastive Learning with Augmented MILPs for Branching (ICLR 2024 · similarity 0.87)
- Rethinking the Capacity of Graph Neural Networks for Branching Strategy (NeurIPS 2024 · similarity 0.87)
- Generalization Guarantees for Learning Score-Based Branch-and-Cut Policies in Integer Programming (NeurIPS 2025 · similarity 0.86)
- Search Strategy Generation for Branch and Bound Using Genetic Programming (AAAI 2025 · similarity 0.86)
- Regularized Large Neighborhood Search (NeurIPS 2026 · similarity 0.86)
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.