Research map / Graph learning / GNN architectures
PipeGCN: Efficient Full-Graph Training of Graph Convolutional Networks with Pipelined Feature Communication
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 |
|---|---|---|---|
| Pipeline inter-partition feature and gradient communication with computation | Distributed full-graph GCN training · Reddit, ogbn-products, Yelp | training throughput: 1.72$\times$$\sim$2.16$\times$ across all datasets and different number of partitions | Helped |
| Pipeline communication without smoothing | Distributed full-graph GCN training · Reddit and Yelp | test score: 97.11$\pm$0.02 -> 97.12$\pm$0.02; 65.26$\pm$0.02 -> 65.27$\pm$0.01 | No clear effect |
| Pipeline communication without smoothing | Distributed full-graph GCN training · ogbn-products | test accuracy: 79.14$\pm$0.35 -> 79.06$\pm$0.42 | Hurt |
| Add feature-gradient smoothing (ablation) | PipeGCN · Reddit (2 partitions) | test accuracy: 97.12$\pm$0.02 -> 97.14$\pm$0.03 | No clear effect |
| Add feature smoothing (ablation) | PipeGCN · ogbn-products (5 partitions) | test accuracy: 79.06$\pm$0.42 -> 79.36$\pm$0.38 | Helped |
| Add feature smoothing (ablation) | PipeGCN · ogbn-products (10 partitions) | test accuracy: 78.91$\pm$0.65 -> 79.21$\pm$0.31 | Helped |
| Add feature and feature-gradient smoothing together (ablation) | PipeGCN · ogbn-products (5 partitions) | test accuracy: 79.06$\pm$0.42 -> 78.86$\pm$0.34 | Hurt |
| Add smoothing to boundary features or feature gradients (ablation) | PipeGCN · training convergence | convergence speed | Helped |
| Use large smoothing decay rates ($0.7$/$0.95$) (ablation) | PipeGCN-GF · ogbn-products | test accuracy over training: $0.7$/$0.95$ | Hurt |
| Use a small smoothing decay rate ($0$ to $0.5$) (ablation) | PipeGCN-GF · ogbn-products | test accuracy: $0$ $\sim$ $0.5$ | Helped |
| Use too-small smoothing decay rate (e.g., 0) (ablation) | PipeGCN-GF · ogbn-products | convergence speed: 0 | Hurt |
| Pipeline training with stale boundary features and feature gradients | PipeGCN · ogbn-papers100M (32 GPUs) | total training time: 38\% | Helped |
Closest papers by meaning
- L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks (CVPR 2020 · similarity 0.94)
- Deeper Insights Into Graph Convolutional Networks for Semi-Supervised Learning (AAAI 2018 · similarity 0.91)
- Simplifying Graph Convolutional Networks (ICML 2019 · similarity 0.90)
- IGLU: Efficient GCN Training via Lazy Updates (ICLR 2022 · similarity 0.90)
- Towards Deepening Graph Neural Networks: A GNTK-based Optimization Perspective (ICLR 2022 · similarity 0.90)
- On the Stability of Graph Convolutional Neural Networks: A Probabilistic Perspective (NeurIPS 2025 · 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.