Research map / Graph learning / GNN architectures
Layer-Neighbor Sampling --- Defusing Neighborhood Explosion in GNNs
Muhammed Fatih Balin, Umit Catalyurek
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 |
|---|---|---|---|
| replace Neighbor Sampling with LABOR-* | GCN sampling · reddit | sampled vertices in layer 3 (K): 167 -> 24 | Improved |
| replace Neighbor Sampling with LABOR-* | GCN sampling · reddit | test F1-score: 96.24 -> 96.23 | No clear effect |
| replace Neighbor Sampling with LABOR-* | GCN sampling · flickr | sampled vertices in layer 3 (K): 73 -> 57 | Improved |
| use LABOR-0 instead of NS for runtime | GCN sampling · reddit | iterations per second: 14.2 -> 37.6 | Improved |
| replace LADIES with Poisson-sampling PLADIES | LADIES · flickr | test F1-score: 50.79 -> 51.52 | Improved |
| replace LADIES with PLADIES | LADIES · reddit | test F1-score: 96.20 -> 96.21 | No clear effect |
| add importance sampling fixed-point iterations (0 to *) (ablation) | LABOR · reddit | sampled vertices in layer 3 (K): 36 -> 24 | Improved |
| add importance sampling fixed-point iterations (0 to *) (ablation) | LABOR · flickr | sampled vertices in layer 3 (K): 66 -> 56 | Improved |
| add importance sampling (LABOR-0 to LABOR-*) (ablation) | LABOR · reddit | edges sampled in layer 2 (K): 177 -> 1070 | Hurt |
| use LABOR-* instead of NS under fixed vertex budget | batch size · reddit | batch size: 90 -> 10100 | Improved |
| tuned LABOR-1 instead of tuned NS for time to target | GCN training · products | time to 91.5% val accuracy (s): 43.82 -> 38.2 | Improved |
| LADIES/PLADIES instead of NS with GATv2 | sampler · reddit | GATv2 runtime | Hurt |
Closest papers by meaning
- Layer-Dependent Importance Sampling for Training Deep and Large Graph Convolutional Networks (NeurIPS 2019 · similarity 0.88)
- Deep Graph Neural Networks via Posteriori-Sampling-based Node-Adaptative Residual Module (NeurIPS 2024 · similarity 0.86)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.86)
- MuseGNN: Forming Scalable, Convergent GNN Layers that Minimize a Sampling-Based Energy (ICLR 2025 · similarity 0.86)
- Integrating Co-Training with Edge Discrimination to Enhance Graph Neural Networks Under Heterophily (AAAI 2025 · similarity 0.86)
- Towards Precise Prediction Uncertainty in GNNs: Refining GNNs with Topology-grouping Strategy (AAAI 2025 · similarity 0.85)
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.
