Research map / Reinforcement learning / Visual and representation RL
Towards Better Branching Policies: Leveraging the Sequential Nature of Branch-and-Bound Tree
Ce Zhang, Bin Zhang, Guoliang Fan
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 contrastive-learning pretraining of the embedding layer (ablation) | Mamba-Branching-p · MILP-S | nodes: 3001 -> 2272 | Improved |
| add contrastive pretraining instead of supervised classification pretraining (ablation) | Mamba-Branching-p · MILP-S | nodes: 4287 -> 2272 | Improved |
| add positional encoding (ablation) | Mamba-Branching-p · MILP-S | nodes: 3481 -> 2272 | Improved |
| use Mamba instead of GRU as sequence model (ablation) | Mamba-Branching-p · MILP-S | nodes: 5684 -> 2272 | Improved |
| use Mamba instead of Transformer as sequence model (ablation) | Mamba-Branching-p · MILP-S | nodes: 5138.15 -> 2272.43 | Improved |
| use sequential history in Mamba instead of no history (TreeGate) | Mamba-Branching · MILP-S | solving time: 113.85 -> 111.78 | Improved |
| use sequential history instead of non-sequential history (T-BranT) | Mamba-Branching · MILP-L easy | solving time: 210.27 -> 123.79 | Improved |
| replace relpscost by Mamba-Branching on difficult instances | branching policy · MILP-L difficult | PD integral: 12741.36 -> 12319.55 | Improved |
| replace relpscost by Mamba-Branching on easy instances | branching policy · MILP-L easy | solving time: 68.43 -> 123.79 | Hurt |
| increase training path length from 100 to 200 steps (ablation) | Mamba-Branching · MILP-S | solving time: 111.78 -> 164.28 | Hurt |
| use relpscost instead of strong branching as imitation expert (ablation) | TreeGate · MILP-S | solving time: 512.32 -> 137.41 | Improved |
| increase inference history from 25 to 100 steps (ablation) | Mamba-Branching · MILP-S | solving time: 111.78 -> 121.98 | Hurt |
Closest papers by meaning
- Generative Branching for Mixed-Integer Linear Programming (AAAI 2026 · similarity 0.92)
- Parameterizing Branch-and-Bound Search Trees to Learn Branching Policies (AAAI 2021 · similarity 0.91)
- CAMBranch: Contrastive Learning with Augmented MILPs for Branching (ICLR 2024 · similarity 0.88)
- Generalization Guarantees for Learning Score-Based Branch-and-Cut Policies in Integer Programming (NeurIPS 2025 · similarity 0.87)
- Rethinking the Capacity of Graph Neural Networks for Branching Strategy (NeurIPS 2024 · similarity 0.87)
- Search Strategy Generation for Branch and Bound Using Genetic Programming (AAAI 2025 · 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.
