Research map / Graph learning / GNN architectures
ProGMLP: A Progressive Framework for GNN-to-MLP Knowledge Distillation with Efficient Trade-offs
Weigang Lu, Ziyu Guan, Wei Zhao, Yaming Yang, Yujie Sun, Zheng Liang, Yibing Zhan, Dapeng Tao
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 |
|---|---|---|---|
| distill GCN teacher into ProGMLP instead of the teacher alone | GCN teacher · Cora/Pubmed/CS/Physics/Photo/Computers (transductive) | accuracy improvement: +1.50% | Improved |
| use E-GLNN ensemble students with GCN teacher | GCN teacher · six transductive datasets | accuracy improvement: -1.67% (E-GLNN2), -1.36% (E-GLNN4) | Hurt |
| use AdaGMLP ensemble students with GCN teacher | GCN teacher · six transductive datasets | accuracy improvement: -1.23% (AdaGMLP2), -0.56% (AdaGMLP4) | Hurt |
| distill into ProGMLP on ogbn-arxiv (GCN teacher) | GCN teacher · ogbn-arxiv | accuracy and inference time: 68.42 -> 70.11 accuracy; 84.13 -> 3.95 ms | Improved |
| distill into ProGMLP on ogbn-products (GCN teacher) | GCN teacher · ogbn-products | accuracy: 72.94 -> 73.00 | No clear effect |
| distill into ProGMLP in the inductive setting (GCN teacher) | GCN teacher · Pubmed (inductive) | accuracy: 80.06 -> 87.95 (+9.85%) | Improved |
| add Progressive Training Structure (initialize each student from the previous one) (ablation) | ProGMLP · Coauthor CS | accuracy: about 4% drop on CS when removed | Improved |
| add Progressive Knowledge Distillation (ablation) | ProGMLP · CS, Physics, Photo, Computers | accuracy | Improved |
| add Progressive Mixup Augmentation (ablation) | ProGMLP · CS, Physics, Photo, Computers | accuracy | Improved |
| vary beta, gamma and early-exit threshold tau_conf (ablation) | ProGMLP · hyperparameter sensitivity | accuracy | No clear effect |
| vary alpha (balance of distillation vs ground-truth loss) (ablation) | ProGMLP · hyperparameter sensitivity | accuracy | Mixed |
| use ProGMLP with GraphSAGE teacher on Physics versus AdaGMLP4 | GraphSAGE teacher · Coauthor Physics | accuracy: 94.20 -> 93.70 | Hurt |
Closest papers by meaning
- Demystifying GNN-to-MLP Knowledge Transfer: Theoretical Grounding and Dual-Stream Distillation Method (AAAI 2026 · similarity 0.91)
- Extracting Low-/High- Frequency Knowledge from Graph Neural Networks and Injecting It into MLPs: An Effective GNN-to-MLP Distillation Framework (AAAI 2023 · similarity 0.91)
- VQGraph: Rethinking Graph Representation Space for Bridging GNNs and MLPs (ICLR 2024 · similarity 0.90)
- TINED: GNNs-to-MLPs by Teacher Injection and Dirichlet Energy Distillation (ICML 2025 · similarity 0.88)
- Large Language Model Meets Graph Neural Network in Knowledge Distillation (AAAI 2025 · similarity 0.88)
- Distilling Graph Geometry: Knowledge Gap from GNNs to MLPs (NeurIPS 2026 · 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.
