Research map / Graph learning / GNN architectures
Full-Graph vs. Mini-Batch Training: Comprehensive Analysis from a Batch Size and Fan-Out Size Perspective
Mengfan Liu, Da Zheng, Junwei Su, Chuan Wu
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| tune batch size and fan-out for mini-batch instead of full-graph | GraphSAGE · Reddit | best test accuracy: 96.13 -> 96.32 | Improved |
| tune batch size and fan-out for mini-batch instead of full-graph | GraphSAGE · ogbn-products | best test accuracy: 77.92 -> 78.80 | Improved |
| tune batch size and fan-out for mini-batch instead of full-graph | GraphSAGE · ogbn-arxiv | best test accuracy: 70.96 -> 71.16 | Improved |
| tune batch size and fan-out for mini-batch instead of full-graph | GraphSAGE · ogbn-papers100M | best test accuracy: 59.54 -> 58.52 | Hurt |
| use iteration-to-accuracy instead of time-to-accuracy across hardware | convergence metric · hardware comparison | max variation across hardware: 2787.05% -> 41.28% | Improved |
| increase batch size with fixed fan-out (MSE) | one-layer GNN · convergence theory | iterations to convergence: T proportional to b^(5/2) | Hurt |
| increase batch size with fixed fan-out (CE) | one-layer GNN · convergence theory | iterations to convergence: T proportional to b^-1 | Improved |
| increase fan-out size under fixed batch size | one-layer GNN · convergence theory | iterations to convergence: T proportional to beta^(-1/2) (MSE), beta^(-5/2) (CE) | Improved |
| increase fan-out size to very large values (over about 15) under CE (ablation) | GraphSAGE · ogbn-products | test accuracy | Hurt |
| increase batch size beyond half of training nodes under CE (ablation) | GraphSAGE · ogbn-products | test accuracy | Hurt |
| use MSE instead of CE for large batch or fan-out | GraphSAGE · ogbn-products | test accuracy | Improved |
Closest papers by meaning
- Bag of Tricks for Training Deeper Graph Neural Networks: A Comprehensive Benchmark Study (TPAMI 2022 · similarity 0.90)
- Measuring the Effects of Data Parallelism on Neural Network Training (JMLR 2019 · similarity 0.89)
- SizeShiftReg: a Regularization Method for Improving Size-Generalization in Graph Neural Networks (NeurIPS 2022 · similarity 0.89)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.88)
- MuseGNN: Forming Scalable, Convergent GNN Layers that Minimize a Sampling-Based Energy (ICLR 2025 · similarity 0.87)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · 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.
