Research map / Reinforcement learning / Visual and representation RL
MIRACLE: Model-free Imitation and Reinforcement Learning for Adaptive Cut-Selection
Arjun M., Rijul Tandon, Agam Gupta, HARIPRASAD KODAMANA, Manojkumar Ramteke
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 weighted curriculum learning (ablation) | MIRACLE · 45 test instances | avg speedup: 1.11× -> 1.34× | Improved |
| add weighted curriculum learning (ablation) | MIRACLE · 45 test instances | success rate: 94% -> 100% | Improved |
| use PPO instead of TRPO (ablation) | MIRACLE · 45 test instances | avg speedup: 1.355× -> 1.338× | No clear effect |
| use PPO instead of SAC (ablation) | MIRACLE · 45 test instances | avg speedup: 1.332× -> 1.338× | No clear effect |
| use 3-layer MLP instead of 2-layer (ablation) | MIRACLE · SetCover | avg speedup: 1.163× -> 1.161× | No clear effect |
| increase expert data from 300 to 1000 instances (ablation) | MIRACLE · SetCover | avg speedup: 1.092× -> 1.093× | No clear effect |
| increase cut budget from 10 to 50 (ablation) | MIRACLE inference · SetCover | avg speedup: 1.170 -> 1.157 | No clear effect |
| increase max iterations from 1 to 10 (ablation) | MIRACLE inference · SetCover | avg speedup: 1.163 -> 1.166 | No clear effect |
| remove early stopping (ablation) | MIRACLE inference · SetCover | avg speedup: 1.166 -> 1.164 | No clear effect |
| use MIRACLE cut selection instead of SCIP default | SCIP 8.0 · SetCover and MIPLIB (300 instances) | peak memory: 2,073.9 MB -> 284.7 MB | Improved |
| use MIRACLE cut selection instead of SCIP baseline | SCIP 8.0 · MIPLIB (Large) | success rate: 53.3% -> 100.0% | Improved |
| use MIRACLE instead of Wang et al. hierarchical model | cut selection · SetCover-Easy | peak memory: ~500 MB -> ~46 MB | Improved |
Closest papers by meaning
- Accelerating Cutting-Plane Algorithms via Reinforcement Learning Surrogates (AAAI 2024 · similarity 0.89)
- Learning to Stop Cut Generation for Efficient Mixed-Integer Linear Programming (AAAI 2024 · similarity 0.86)
- Learning to Remove Cuts in Integer Linear Programming (ICML 2024 · similarity 0.85)
- ART for Diffusion Sampling: A Reinforcement Learning Approach to Timestep Scheduling (NeurIPS 2026 · similarity 0.85)
- Reinforcement Learning with Multi-Step Lookahead Information Via Adaptive Batching (NeurIPS 2026 · similarity 0.85)
- ESSAM: A Novel Competitive Evolution Strategies Approach to Reinforcement Learning for Memory Efficient LLMs Fine-Tuning (NeurIPS 2026 · similarity 0.84)
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.
