Research map / Graph learning / Graph algorithms and clustering
A General Neural Backbone for Mixed-Integer Linear Optimization via Dual Attention
Peixin Huang, Yaoxin Wu, Yining Ma, Cathy Wu, Wei Zhang, Wen Song
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 inter-type cross-attention (ablation) | Dual-attention backbone · Binary variable prediction, WA | macro F1: 0.7860 -> 0.9205 | Improved |
| add intra-type self-attention (ablation) | Dual-attention backbone · Binary variable prediction, WA | macro F1: 0.8992 -> 0.9205 | Improved |
| add inter-type cross-attention (ablation) | Dual-attention backbone · Binary variable prediction, WA | error rate: 14.18% -> 5.32% | Improved |
| use GATv2 attention in BGNN (BGAT) (ablation) | BGNN · Binary variable prediction, WA | macro F1: 0.8986 -> 0.8973 | No clear effect |
| add self-attention to BGNN message passing (ablation) | BGNN · Binary variable prediction, WA | macro F1: 0.8986 -> 0.9017 | No clear effect |
| replace BGNN with dual-attention backbone | BGNN · Binary variable prediction, WA | macro F1: 0.8986 -> 0.9205 | Improved |
| replace BGNN with dual-attention backbone | BGNN · Binary variable prediction, IP | macro F1: 0.5060 -> 0.5089 | Improved |
| replace BGNN with dual-attention backbone | BGNN · Feasibility prediction, D2 foldable | feasibility error rate: 6.0% -> 0.0% | Improved |
| replace BGNN with dual-attention backbone in PaS | PaS-BGNN · Predict-and-search, IP | primal gap: 7.435% -> 5.556% | Improved |
| replace BGNN with dual-attention backbone in Apollo | Apollo-BGNN · Predict-and-search, IP | primal gap: 4.984% -> 4.167% | Improved |
| replace BGNN with dual-attention backbone in learning-to-branch | BGNN · B&B tree size, MIS transfer | nodes: 10280.5 -> 8938.5 | Improved |
| replace BGNN with dual-attention backbone on MIPLIB WA | BGNN · MIPLIB WA | MSE: 0.2412 -> 0.5174 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Polynormer — Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Expressive Power of Graph Neural Networks for (Mixed-Integer) Quadratic Programs (ICML 2025 · similarity 0.87)
- BIPNN: Learning to Solve Binary Integer Programming via Hypergraph Neural Networks (NeurIPS 2025 · similarity 0.87)
- On the Power of Small-size Graph Neural Networks for Linear Programming (NeurIPS 2024 · similarity 0.86)
- Are Graph Neural Networks Optimal Approximation Algorithms? (NeurIPS 2024 · similarity 0.86)
- Apollo-MILP: An Alternating Prediction-Correction Neural Solving Framework for Mixed-Integer Linear Programming (ICLR 2025 · similarity 0.86)
- Geometric Algorithms for Neural Combinatorial Optimization with Constraints (NeurIPS 2025 · similarity 0.86)
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.
