Research map / Reinforcement learning / Visual and representation RL
Accelerate Presolve in Large-Scale Linear Programming via Reinforcement Learning
Yufei Kuang, Xijun Li, Jie Wang, Fangzhou Zhu, Meng Lu, Zhihai Wang, Jia Zeng, Houqiang Li, Yongdong Zhang, Feng Wu
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 |
|---|---|---|---|
| replace default presolve routine with RL4Presolve | Clp · Master Production Schedule | solving time (s): 2.65 -> 1.58 | Improved |
| replace default presolve routine with RL4Presolve | Clp · Production Planning | solving time (s): 6.15 -> 4.26 | Improved |
| replace default presolve routine with RL4Presolve | Clp · Set Covering | solving time (s): 16.64 -> 1.94 | Improved |
| replace default presolve routine with RL4Presolve | Clp · MIRPLIB | solving time (s): 126.37 -> 107.35 | Improved |
| replace default presolve routine with RL4Presolve | OptVerse · Multicommodity Network Flow | solving time (s): 2.24 -> 1.36 | Improved |
| apply policy trained on dual simplex to primal simplex | RL4Presolve · Production Planning | solving time (s): 7.72 -> 5.49 | Improved |
| apply policy trained on dual simplex to interior point | RL4Presolve · Production Planning | solving time (s): 130.47 -> 137.39 | Hurt |
| apply policy to larger instances | RL4Presolve · Production Planning (Large) | solving time (s): 51.17 -> 40.27 | Improved |
| use rules extracted from learned policies instead of default routine | Clp · Production Planning | solving time (s): 6.15 -> 4.40 | Improved |
| use adaptive action sequences instead of single-presolver actions (ablation) | RL4Presolve · Master Production Schedule | total decision time (s): 6.78 -> 0.26 | Improved |
| disable presolve | Clp default routine · Production Planning | solving time (s): 4.97 -> 45.21 | Hurt |
| keep only last 60% presolvers by NNZ reduction | Clp default routine · Production Planning | solving time (s): 4.97 -> 27.42 | Hurt |
Closest papers by meaning
- Solver-Informed RL: Grounding Large Language Models for Authentic Optimization Modeling (NeurIPS 2025 · similarity 0.87)
- Heuristic-Guided Reinforcement Learning (NeurIPS 2021 · similarity 0.87)
- MetaBox-v2: A Unified Benchmark Platform for Meta-Black-Box Optimization (NeurIPS 2025 · similarity 0.86)
- Think Twice, Act Once: A Co-Evolution Framework of LLM and RL for Large-Scale Decision Making (ICML 2025 · similarity 0.85)
- SRL: Scaling Distributed Reinforcement Learning to Over Ten Thousand Cores (ICLR 2024 · similarity 0.85)
- Planning, Fast and Slow: Online Reinforcement Learning with Action-Free Offline Data via Multiscale Planners (ICML 2024 · 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.
