Research map / Graph learning / GNN architectures
Layer-Dependent Importance Sampling for Training Deep and Large Graph Convolutional Networks
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 layer-dependent importance sampling instead of FastGCN sampling | 5-layer GCN · Cora | F1-Score(%): 78.0 ± 2.1 -> 78.3 ± 1.6 | Helped |
| Use layer-dependent importance sampling instead of FastGCN sampling | 5-layer GCN · Citeseer | F1-Score(%): 63.5 ± 1.8 -> 65.0 ± .1.4 | Helped |
| Use layer-dependent importance sampling instead of FastGCN sampling | 5-layer GCN · Pubmed | F1-Score(%): 74.4 ± 0.8 -> 76.8 ± 0.8 | Helped |
| Use layer-dependent importance sampling instead of GraphSage neighborhood sampling | 5-layer GCN · Reddit | F1-Score(%): 92.1 ± 1.1 -> 92.8 ± 1.6 | Helped |
| Restrict candidate samples to neighbors of nodes sampled in the upper layer | LADIES · GCN embedding estimation (theoretical analysis) | estimation variance | Helped |
| Use layer-wise sampling rather than recursively sampling neighbors per node | LADIES · Deep GCN training (theoretical complexity analysis) | time complexity | Helped |
| Reduce LADIES sample count from 512 to 64 | 5-layer GCN · Cora | F1-Score(%): 78.3 ± 1.6 -> 77.6 ± 1.4 | No clear effect |
| Reduce LADIES sample count from 512 to 64 | 5-layer GCN · Citeseer | F1-Score(%): 64.3 ± 2.4 -> 65.0 ± .1.4 | Helped |
| Reduce LADIES sample count from 512 to 64 | 5-layer GCN · Pubmed | F1-Score(%): 75.9 ± 1.1 -> 76.8 ± 0.8 | Helped |
| Reduce LADIES sample count from 512 to 64 | 5-layer GCN · Reddit | F1-Score(%): 92.8 ± 1.6 -> 83.5 ± 0.9 | Hurt |
| Use stochastic LADIES sampling rather than full-batch training | GCN · Pubmed validation and test sets | F1-Score | Helped |
Closest papers by meaning
- FastGCN: Fast Learning with Graph Convolutional Networks via Importance Sampling (ICLR 2018 · similarity 0.92)
- L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks (CVPR 2020 · similarity 0.92)
- Generalization Guarantee of Training Graph Convolutional Networks with Graph Topology Sampling (ICML 2022 · similarity 0.92)
- Adaptive Sampling Towards Fast Graph Representation Learning (NeurIPS 2018 · similarity 0.91)
- Stochastic Training of Graph Convolutional Networks with Variance Reduction (ICML 2018 · similarity 0.91)
- GraphSAINT: Graph Sampling Based Inductive Learning Method (ICLR 2020 · similarity 0.91)
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.