Research map / Large language models / Agents, code and math
LLM4Branch: Large Language Model for Discovering Efficient Branching Policies of Integer Programs
Zhinan Hou, Xingchen Li, Yankai Zhang, Tianxun Li, Keyou You
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 |
|---|---|---|---|
| optimize solver cost instead of strong-branching imitation accuracy (ablation) | LLM4Branch · Setcover (Easy) | solving time (s): 3.88 -> 3.21 | Improved |
| optimize solver cost instead of strong-branching imitation accuracy (ablation) | LLM4Branch · Indset (Easy) | solving time (s): 45.21 -> 10.91 | Improved |
| add parameter optimization stage (ablation) | LLM4Branch · Setcover (Easy) | solving time (s): 4.53 -> 3.21 | Improved |
| add parameter optimization stage (ablation) | LLM4Branch · Indset (Easy) | solving time (s): 28.09 -> 10.91 | Improved |
| add natural-language feature description (prior) to prompt (ablation) | LLM4Branch · Setcover (Easy) | solving time (s): 4.10 -> 3.21 | Improved |
| add natural-language feature description (prior) to prompt (ablation) | LLM4Branch · Indset (Easy) | solving time (s): 17.03 -> 10.91 | Improved |
| change LLM backbone from DeepSeek-R1 to GPT-5, DeepSeek-V3 or Qwen3-Next-80B-A3B (ablation) | LLM4Branch · Setcover (Easy) | solving time (s): 3.21 vs 3.25, 3.39, 3.55 | No clear effect |
| change LLM backbone from DeepSeek-R1 to GPT-5, DeepSeek-V3 or Qwen3-Next-80B-A3B (ablation) | LLM4Branch · Indset (Easy) | solving time (s): 10.91 vs 9.01, 9.34, 10.52 | No clear effect |
| select MLP checkpoint by solving time rather than loss or accuracy (ablation) | MLP policy · Indset (500 nodes, affinity 4) | solve time (s): 3.047 -> 3.016 | Improved |
| select MLP checkpoint by highest accuracy rather than lowest loss (ablation) | MLP policy · Indset (500 nodes, affinity 4) | solve time (s): 3.047 -> 3.044 | No clear effect |
| use LLM-discovered policy instead of GNN imitating strong branching | LLM4Branch · Indset (Easy) | strong-branching accuracy (%): 48.84 -> 24.72 | Hurt |
| use LLM4Branch instead of Hybrid | Hybrid · Indset (Medium) | solving time (s): 112.48 -> 62.45 | Improved |
Closest papers by meaning
- CAMBranch: Contrastive Learning with Augmented MILPs for Branching (ICLR 2024 · similarity 0.88)
- 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 0.87)
- Large Language Model-driven Large Neighborhood Search for Large-Scale MILP Problems (ICML 2025 · similarity 0.87)
- OptiBench Meets ReSocratic: Measure and Improve LLMs for Optimization Modeling (ICLR 2025 · similarity 0.86)
- Predicting Lagrangian Multipliers for Mixed Integer Linear Programs (ICML 2024 · 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.
