Research map / Graph learning / GNN architectures
FastGCN: Fast Learning with Graph Convolutional Networks via Importance Sampling
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 |
|---|---|---|---|
| Train with FastGCN vertex sampling rather than batched GCN | GCN · Reddit | total training time: 638.6 vs 58346.6 seconds | Helped |
| Train with FastGCN vertex sampling rather than GraphSAGE-GCN | GCN · Reddit | total training time: 638.6 vs 3318.5 seconds | Helped |
| Train with FastGCN vertex sampling rather than batched GCN | GCN · Pubmed | total training time: 15.5 vs 210.8 seconds | Helped |
| Train with FastGCN vertex sampling rather than GraphSAGE-GCN | GCN · Cora | total training time: 2.7 vs 72.4 seconds | Helped |
| Use inductive FastGCN rather than FastGCN-transductive | FastGCN · Cora, original split | F1: 0.723 vs 0.818 | Hurt |
| Use inductive FastGCN rather than FastGCN-transductive | FastGCN · Pubmed, original split | F1: 0.721 vs 0.776 | Hurt |
| Use importance sampling instead of uniform sampling (ablation) | FastGCN · Cora, Pubmed, and Reddit | prediction accuracy | Helped |
| Increase the number of sampled vertices (ablation) | FastGCN · Cora, Pubmed, and Reddit | micro F1 | Helped |
| Precompute the bottom-layer product instead of sampling that layer (ablation) | FastGCN · Cora, Pubmed, and Reddit | training time | Helped |
| Precompute the bottom-layer product instead of sampling that layer (ablation) | FastGCN · Cora, Pubmed, and Reddit | prediction accuracy | No clear effect |
| Use the original Cora and Pubmed split with fewer training labels (ablation) | FastGCN · Cora and Pubmed | F1 compared with GCN: Cora: 0.723 vs 0.814; Pubmed: 0.721 vs 0.795 | Mixed |
Closest papers by meaning
- Layer-Dependent Importance Sampling for Training Deep and Large Graph Convolutional Networks (NeurIPS 2019 · similarity 0.92)
- L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks (CVPR 2020 · similarity 0.92)
- Deeper Insights Into Graph Convolutional Networks for Semi-Supervised Learning (AAAI 2018 · similarity 0.91)
- Simplifying Graph Convolutional Networks (ICML 2019 · similarity 0.91)
- GCN meets GPU: Decoupling “When to Sample” from “How to Sample” (NeurIPS 2020 · similarity 0.90)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · similarity 0.89)
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.