Research map / Graph learning / GNN architectures
GNNAutoScale: Scalable and Expressive Graph Neural Networks via Historical Embeddings
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 |
|---|---|---|---|
| Use GAS historical embeddings rather than full-batch execution | GCN, GAT, APPNP, and GCNII · Small transductive benchmarks | mean accuracy difference (percentage points): GCN +0.13; GAT +0.29; APPNP -0.01; GCNII +0.29 | No clear effect |
| Add Metis clustering to reduce inter-batch connectivity (ablation) | GAS GCNII · Cora, CiteSeer, PubMed, Coauthor-CS, Coauthor-Physics, Amazon-Computer, Amazon-Photo, Wiki-CS | accuracy difference from full-batch (percentage points): Baseline -3.26; Metis -1.57 (Cora) | Helped |
| Add regularization (ablation) | GAS GCNII · Cora, CiteSeer, PubMed, Coauthor-CS, Coauthor-Physics, Amazon-Computer, Amazon-Photo, Wiki-CS | accuracy difference from full-batch (percentage points): Baseline -3.26; Regularization -2.12 (Cora) | Helped |
| Use both Metis clustering and regularization (ablation) | GAS GCNII · Cora, CiteSeer, PubMed, Coauthor-CS, Coauthor-Physics, Amazon-Computer, Amazon-Photo, Wiki-CS | accuracy difference from full-batch (percentage points): Cora +0.48; CiteSeer +0.83; PubMed +0.47; Coauthor-CS +0.07; Coauthor-Physics +0.18; Amazon-Computer +0.01; Amazon-Photo | Helped |
| Add Metis clustering (ablation) | GAS GIN · Cluster test | accuracy: 55.15 -> 57.82 | Helped |
| Add Lipschitz continuity regularization (ablation) | GAS GIN with Metis clustering · Cluster test | accuracy: 57.82 -> 58.51 | Helped |
| Add Lipschitz continuity regularization (ablation) | GAS with linear GCN operators · GCN | accuracy | No clear effect |
| Replace GTTF with GAS | 4-layer GCN · Cora | runtime (s): 0.077 -> 0.006 | Helped |
| Replace GTTF with GAS | 4-layer GCN · Cora | memory (MB): 18.01 -> 2.13 | Helped |
| Use Metis partitions instead of random mini-batches (ablation) | GAS batch selection · Real-world graph datasets | inter-/intra-connectivity ratio: factor of 4 on average | Helped |
| Use GAS with deep or expressive GNNs instead of the shallow GCN+GAS baseline | GCNII and PNA · Reddit, PPI, Flickr, Yelp, ogbn-arxiv, ogbn-products | accuracy: Reddit: 95.45 -> 96.77 (GCNII), 95.45 -> 97.17 (PNA); Flickr: 54.00 -> 56.20 (GCNII), 54.00 -> 56.67 (PNA); Yelp: 62.94 | Mixed |
Closest papers by meaning
- Haste Makes Waste: A Simple Approach for Scaling Graph Neural Networks (ICML 2025 · similarity 0.90)
- Neural Snowflakes: Universal Latent Graph Inference via Trainable Latent Geometries (ICLR 2024 · similarity 0.88)
- MuseGNN: Forming Scalable, Convergent GNN Layers that Minimize a Sampling-Based Energy (ICLR 2025 · similarity 0.88)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · similarity 0.87)
- Generative Graph Pattern Machine (NeurIPS 2025 · similarity 0.87)
- Towards characterizing the value of edge embeddings in Graph Neural Networks (ICML 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.