Research map / Graph learning / Knowledge and temporal graphs
Learning Large Graph Property Prediction via Graph Segment Training
Kaidi Cao, Phitchaya Mangpo Phothilimthana, Sami Abu-El-Haija, Dustin Zelle, Yanqi Zhou, Charith Mendis, Jure Leskovec, Bryan Perozzi
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 |
|---|---|---|---|
| train with graph segments instead of full graph | GCN · MalNet-Tiny | test accuracy: 87.84 -> 88.26 | No clear effect |
| train on only one random segment per graph (GST-One) (ablation) | GST with SAGE · MalNet-Large | test accuracy: 88.62 -> 57.13 | Hurt |
| add historical embedding table alone (GST+E) (ablation) | GST with GCN · MalNet-Large | test accuracy: 88.35 -> 48.42 | Hurt |
| add prediction head finetuning to GST+E (ablation) | GST+E with SAGE · MalNet-Large | test accuracy: 43.28 -> 85.26 | Improved |
| add stale embedding dropout (SED) to GST+E (ablation) | GST+E with SAGE · MalNet-Large | test accuracy: 43.28 -> 71.83 | Improved |
| add SED to GST+EF (full GST+EFD) (ablation) | GST+EF with SAGE · MalNet-Large | test accuracy: 85.26 -> 89.78 | Improved |
| GST+EFD instead of GST | GCN · MalNet-Large | test accuracy: 88.35 -> 89.67 | Improved |
| GST+EFD instead of GST | GraphGPS · MalNet-Tiny | test accuracy: 91.03 -> 92.46 | Improved |
| GST+EFD instead of GST | TpuGraphs · TpuGraphs | test OPA: 86.84 -> 89.20 | Improved |
| GST+EFD instead of GST: lower train OPA | TpuGraphs · TpuGraphs | train OPA: 87.43 -> 71.40 | Hurt |
| GST+EFD instead of GST for runtime | GCN · MalNet-Large | training time per iteration (ms): 720.8 -> 247.9 | Improved |
| SED embedding fetch overhead: GST+EFD vs GST-One runtime | GCN · MalNet-Large | training time per iteration (ms): 242.2 -> 247.9 | No clear effect |
Closest papers by meaning
- GraphZoom: A Multi-level Spectral Approach for Accurate and Scalable Graph Embedding (ICLR 2020 · similarity 0.87)
- Haste Makes Waste: A Simple Approach for Scaling Graph Neural Networks (ICML 2025 · similarity 0.86)
- Two Heads Are Better Than One: Boosting Graph Sparse Training via Semantic and Topological Awareness (ICML 2024 · similarity 0.85)
- Graph Classification via Reference Distribution Learning: Theory and Practice (NeurIPS 2024 · similarity 0.85)
- Sketch-Augmented Features Improve Learning Long-Range Dependencies in Graph Neural Networks (NeurIPS 2025 · similarity 0.85)
- Neural Snowflakes: Universal Latent Graph Inference via Trainable Latent Geometries (ICLR 2024 · 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.
