Research map / Reinforcement learning / Visual and representation RL
Promoting Generalization for Exact Combinatorial Solvers via Adversarial Instance Augmentation
Haoyang Liu, Jie Wang, Yufei Kuang, Xijun 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 |
|---|---|---|---|
| add adversarial instance augmentation to GNN branching (AdaSolver-IL) | GNN · Set Covering D1 | solving time: 9.25 -> 6.28 | Improved |
| add adversarial instance augmentation to GNN branching (AdaSolver-IL) | GNN · easier distributions, four benchmarks | solving time: 24.60% improvement of solving time | Improved |
| add adversarial instance augmentation to tMDP+DFS (AdaSolver-RL) | tMDP+DFS · easier distributions, four benchmarks | solving time: 10.75% improvement | Improved |
| add adversarial instance augmentation to GNN branching on larger instances | GNN · harder distributions D4-D6 | PD integral: 9.53% over GNN and 11.68% over tMDP+DFS | Improved |
| add AdaSolver-IL on real-world Anonymous dataset | GNN · Anonymous (MIRPLIB) | PD gap: 1.10 -> 0.92 | Improved |
| train GNN on 100 instead of 10000 instances | GNN · four benchmarks | solving time: 115.58% increase in solving time and 25.57% increase in PD integral | Hurt |
| add AdaSolver-IL with only 100 training instances | GNN (100 instances) · low-data training | solving time: 36.20% improvement in solving time and 16.94% in PD integral | Improved |
| replace adversarial augmentation policy with random masking (IL), i.e. adding adversarial policy (ablation) | AdaSolver-IL · Set Covering D4 | PD integral: 15550.73 -> 10691.05 | Improved |
| add adversarial policy instead of random masking (IL) (ablation) | AdaSolver-IL · Set Covering D1 | solving time: 6.13 vs 6.28 | No clear effect |
| add adversarial policy instead of random masking (RL) (ablation) | AdaSolver-RL · Capacitated Facility Location D4 | PD integral: 8579.21 -> 7779.28 | Improved |
| raise masking proportion from 0.01 to 0.05 (ablation) | AdaSolver-IL · Set Covering D1 | solving time: 6.28 -> 9.20 | Hurt |
| set beta to 0.1 instead of 0.01 (ablation) | AdaSolver-IL · Combinatorial Auctions D2 | solving time: 16.34 -> 23.84 | Hurt |
Closest papers by meaning
- Generalization of Neural Combinatorial Solvers Through the Lens of Adversarial Robustness (ICLR 2022 · similarity 0.87)
- Principled Data Augmentation for Learning to Solve Quadratic Programming Problems (NeurIPS 2025 · similarity 0.86)
- NeuralSolver: Learning Algorithms For Consistent and Efficient Extrapolation Across General Tasks (NeurIPS 2024 · similarity 0.86)
- RoME: Domain-Robust Mixture-of-Experts for MILP Solution Prediction across Domains (NeurIPS 2025 · similarity 0.86)
- GOAL: A Generalist Combinatorial Optimization Agent Learner (ICLR 2025 · similarity 0.85)
- Neural Solver Selection for Combinatorial Optimization (ICML 2025 · 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.
