Research map / Graph learning / GNN architectures
MLPInit: Embarrassingly Simple GNN Training Acceleration with MLP Initialization
Xiaotian Han, Tong Zhao, Yozen Liu, Xia Hu, Neil Shah
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 |
|---|---|---|---|
| Initialize GNNs with converged PeerMLP weights and fine-tune | GraphSAGE node classification · 7 node-classification datasets | average accuracy improvement: +7.97% | Improved |
| Initialize GNNs with converged PeerMLP weights and fine-tune | GraphSAINT node classification · 7 node-classification datasets | average accuracy improvement: +7.00% | Improved |
| Initialize GNNs with converged PeerMLP weights and fine-tune | ClusterGCN node classification · 7 node-classification datasets | average accuracy improvement: +6.61% | Improved |
| Initialize GNNs with converged PeerMLP weights and fine-tune | GCN node classification · 7 node-classification datasets | average accuracy improvement: +14.00% average; OGB-products: 77.08 -> 76.85 | Mixed |
| Initialize GraphSAGE with converged PeerMLP weights | GraphSAGE · OGB-products | training speedup: 14.83× | Improved |
| Initialize GCN with converged PeerMLP weights | GCN · OGB-products | training speedup: 33.69× | Improved |
| Initialize GraphSAGE with converged PeerMLP weights | GraphSAGE link prediction · PubMed | Hits@10: 28.38 -> 37.58 | Improved |
| Initialize GCN with converged PeerMLP weights | GCN link prediction · A-Photo | Hits@10: 9.63 -> 9.17 | Hurt |
| Initialize GCN with converged PeerMLP weights | GCN link prediction · A-Computers | Hits@10: 7.60 -> 6.76 | Hurt |
| Use PeerMLP weights without fine-tuning the GNN | GCN · Cora node classification | accuracy: 58.50 -> 77.60 | Improved |
| Use PeerMLP weights without fine-tuning the GNN | GCN · CiteSeer node classification | accuracy: 60.50 -> 69.70 | Improved |
| Use PeerMLP weights without fine-tuning the GNN | GCN · PubMed node classification | accuracy: 73.60 -> 78.10 | Improved |
Closest papers by meaning
- Graph-less Neural Networks: Teaching Old MLPs New Tricks Via Distillation (ICLR 2022 · similarity 0.89)
- Scalable Graph Neural Networks via Bidirectional Propagation (NeurIPS 2020 · similarity 0.89)
- TINED: GNNs-to-MLPs by Teacher Injection and Dirichlet Energy Distillation (ICML 2025 · similarity 0.89)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · similarity 0.87)
- VQGraph: Rethinking Graph Representation Space for Bridging GNNs and MLPs (ICLR 2024 · similarity 0.86)
- Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification (NeurIPS 2024 · 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.
