Research map / Graph learning / GNN architectures
Extracting Low-/High- Frequency Knowledge from Graph Neural Networks and Injecting It into MLPs: An Effective GNN-to-MLP Distillation Framework
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace node-to-node GLNN distillation with FF-G2M full-frequency distillation | MLP student with GCN teacher · Cora | accuracy (%): 82.8 -> 84.3 | Helped |
| replace node-to-node GLNN distillation with FF-G2M full-frequency distillation | MLP student with GCN teacher · Citeseer | accuracy (%): 72.7 -> 74.0 | Helped |
| replace node-to-node GLNN distillation with FF-G2M full-frequency distillation | MLP student with GCN teacher · Pubmed | accuracy (%): 80.2 -> 81.8 | Helped |
| replace node-to-node GLNN distillation with FF-G2M full-frequency distillation | MLP student with GCN teacher · Amazon-Photo | accuracy (%): 91.4 -> 94.2 | Helped |
| replace node-to-node GLNN distillation with FF-G2M full-frequency distillation | MLP student with GCN teacher · Coauthor-CS | accuracy (%): 92.7 -> 93.8 | Helped |
| replace node-to-node GLNN distillation with FF-G2M full-frequency distillation | MLP student with GCN teacher · Coauthor-Phy | accuracy (%): 93.2 -> 94.4 | Helped |
| add low-frequency neighborhood distillation (LFD) (ablation) | MLP student with GCN teacher · six graph datasets | accuracy (%): 83.4 ± 0.9; 73.7 ± 0.6; 81.0 ± 0.5; 92.1 ± 0.8; 93.2 ± 0.8; 93.7 ± 0.8 | Helped |
| add high-frequency pairwise-difference distillation (HFD) alone (ablation) | vanilla MLP with GCN teacher · six graph datasets | accuracy (%): 68.5 ± 0.8; 63.2 ± 0.7; 74.4 ± 0.4; 82.5 ± 1.3; 89.3 ± 1.7; 91.0 ± 1.6 | Helped |
| add HFD to LFD (ablation) | MLP student with GCN teacher · six graph datasets | accuracy (%): 83.4 ± 0.9 -> 84.3 ± 0.4; 73.7 ± 0.6 -> 74.0 ± 0.5; 81.0 ± 0.5 -> 81.8 ± 0.4; 92.1 ± 0.8 -> 94.2 ± 0.4; 93.2 ± 0.8 -> 93 | Helped |
Closest papers by meaning
- VQGraph: Rethinking Graph Representation Space for Bridging GNNs and MLPs (ICLR 2024 · similarity 0.91)
- TINED: GNNs-to-MLPs by Teacher Injection and Dirichlet Energy Distillation (ICML 2025 · similarity 0.90)
- Distilling Graph Geometry: Knowledge Gap from GNNs to MLPs (NeurIPS 2026 · similarity 0.89)
- Large Language Model Meets Graph Neural Network in Knowledge Distillation (AAAI 2025 · similarity 0.88)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.88)
- Graph Distillation with Eigenbasis Matching (ICML 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.