Research map / Graph learning / Graph algorithms and clustering
MILPnet: A Multi-Scale Architecture with Geometric Feature Sequence Representations for Advancing MILP Problems
Ruobing Wang, Xin Li, Mingzhong Wang
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 hybrid attention (HYA) (ablation) | MILPnet · FOLD(20,6) feasibility | MSE: 0.2503 -> 0.0001 | Improved |
| add multi-scale attention (MSA) (ablation) | MILPnet · FOLD(20,6) feasibility | MSE: 0.2500 -> 0.0001 | Improved |
| add multi-scale attention (MSA) (ablation) | MILPnet · FOLD(100,20) feasibility | ErrorN: 4995 -> 6 | Improved |
| max window n/2 instead of n/4 (ablation) | MILPnet · FOLD(50,30) feasibility | MSE: 0.0841 -> 0.0477 | Improved |
| max window n/4 vs n/2 (ablation) | MILPnet · FOLD(20,16) | MSE: 0.0005 -> 0.0026 | Mixed |
| increase attention blocks from 1 to 2 (ablation) | MILPnet · FOLD(20,16) feasibility | MSE: 0.0006 -> 0.0001 | Improved |
| increase attention blocks from 2 to 3 (ablation) | MILPnet · FOLD(20,16) feasibility | MSE: 0.0001 -> 0.0004 | No clear effect |
| randomly permute constraint order (ablation) | MILPnet · FOLD(50,20) feasibility | MSE: 0.0005 -> 0.0006 | No clear effect |
| sequence model instead of GCN | GCN · FOLD(50,30) feasibility | MSE: 0.3556 -> 0.0023 | Improved |
| sequence model instead of graph models | SAGE · FOLD(50,20) optimal solution | MSE: 0.1000 -> 0.1158 | Hurt |
| sequence model instead of GCN | GCN · FOLD(20,6) optimal solution | MSE: 0.0751 -> 0.0473 | Improved |
| random feature augmentation | GCN · FOLD(50,30) feasibility | MSE: 0.3556 -> 0.5223 | Hurt |
Closest papers by meaning
- Rethinking the Capacity of Graph Neural Networks for Branching Strategy (NeurIPS 2024 · similarity 0.89)
- MILP-StuDio: MILP Instance Generation via Block Structure Decomposition (NeurIPS 2024 · similarity 0.88)
- Towards Foundation Models for Mixed Integer Linear Programming (ICLR 2025 · similarity 0.88)
- RoME: Domain-Robust Mixture-of-Experts for MILP Solution Prediction across Domains (NeurIPS 2025 · similarity 0.88)
- Apollo-MILP: An Alternating Prediction-Correction Neural Solving Framework for Mixed-Integer Linear Programming (ICLR 2025 · similarity 0.87)
- On the Power of Small-size Graph Neural Networks for Linear Programming (NeurIPS 2024 · 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.
