Research map / Reinforcement learning / Imitation learning
Reinforcement Learning for Branch-and-Bound Optimisation Using Retrospective Trajectories
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 retrospectively constructed within-subtree trajectories instead of full-episode trajectories (ablation) | RL branching on Set Covering (500×1000) · Set Covering (500×1000) | # B&B nodes: 14× fewer nodes | Helped |
| Use retrospective trajectories rather than FMSTS-style training without retrospective trajectories (ablation) | RL branching on Set Covering (500×1000) · Set Covering (500×1000), DFS test setting | # B&B nodes: 5× | Helped |
| Use retrospective trajectories rather than FMSTS-style training without retrospective trajectories (ablation) | RL branching on Set Covering (500×1000) · Set Covering (500×1000), non-DFS test setting | # B&B nodes: 3× | Helped |
| Use a non-DFS node selection strategy instead of DFS during training and evaluation (ablation) | FMSTS · Set Covering (500×1000) | # B&B nodes | Helped |
| Use maximum LP gain rather than random, visitation-order, or deepest trajectory construction (ablation) | Retro · Set Covering (500×1000) | # B&B nodes | No clear effect |
| Use retro branching rather than pseudocost branching | Branching policy · Set Covering, Combinatorial Auction, Capacitated Facility Location, Maximum Independent Set | # Nodes | Mixed |
| Use retro branching rather than imitation learning | Branching policy · Set Covering (small) | # Nodes: 10.5 -> 8.68 | Helped |
| Use retro branching rather than imitation learning | Branching policy · Combinatorial Auction | # Nodes: 5.29 -> 5.73 | Hurt |
| Use retro branching rather than imitation learning | Branching policy · Capacitated Facility Location | # Nodes: 10.2 -> 10.1 | Helped |
| Use retro branching rather than imitation learning | Branching policy · Maximum Independent Set | # Nodes: 4.08 -> 4.01 | Helped |
| Use retro branching rather than imitation learning | Branching policy · Set Covering (500×1000) | # B&B tree size: ≈20% larger | Hurt |
| Train the Retro agent with full-episode trajectories rather than retrospective trajectories (ablation) | RL branching · Set Covering (500×1000) | # B&B nodes: 14× fewer nodes with Retro | Hurt |
Closest papers by meaning
- Planning in Branch-and-Bound: Model-Based Reinforcement Learning for Exact Combinatorial Optimization (AAAI 2026 · similarity 0.93)
- SORREL: Suboptimal-Demonstration-Guided Reinforcement Learning for Learning to Branch (AAAI 2025 · similarity 0.88)
- CAMBranch: Contrastive Learning with Augmented MILPs for Branching (ICLR 2024 · similarity 0.88)
- Reinforcement learning with combinatorial actions for coupled restless bandits (ICLR 2025 · similarity 0.88)
- Preference Optimization for Combinatorial Optimization Problems (ICML 2025 · similarity 0.87)
- Structured Reinforcement Learning for Combinatorial Decision-Making (NeurIPS 2025 · similarity 0.87)
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.