Research map / Graph learning / Graph self-supervised learning
Large-Scale Representation Learning on Graphs via Bootstrapping
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 BGRL instead of GRACE | 2-layer GCN · Amazon Computers | classification accuracy: 89.53 ± 0.35 -> 90.34 ± 0.19 | Helped |
| Use BGRL instead of GRACE | 2-layer GCN · WikiCS | classification accuracy: 80.14 ± 0.48 -> 79.98 ± 0.10 | No clear effect |
| Use BGRL instead of GRACE | 3-layer GCN · ogbn-arXiv test | classification accuracy: 70.33 ± 0.18 -> 71.64 ± 0.12 | No clear effect |
| Use BGRL rather than GRACE with a GAT encoder | PPI · PPI | Micro-F1: 69.71 ± 0.17 -> 70.49 ± 0.05 | Helped |
| Replace standard augmentation with degree-centrality augmentation (ablation) | BGRL · Coauthor CS | classification accuracy: 93.31 ± 0.13 -> 93.34 ± 0.13 | No clear effect |
| Replace standard augmentation with pagerank-centrality augmentation (ablation) | BGRL · Amazon Computers | classification accuracy: 90.34 ± 0.19 -> 90.45 ± 0.25 | No clear effect |
| Add a projector network (ablation) | BGRL on MAG240M · MAG240M | classification accuracy | Hurt |
| Increase the number of sampled negatives from k=2 to k=8 (ablation) | GRACE-SUBSAMPLING · ogbn-arXiv test | classification accuracy: 60.24 ± 4.06 -> 70.33 ± 0.18 | Helped |
| Increase the number of sampled negatives from k=8 to k=32 (ablation) | GRACE-SUBSAMPLING · ogbn-arXiv test | classification accuracy: 70.33 ± 0.18 -> 71.18 ± 0.16 | Helped |
| Add unlabeled nodes to each training minibatch for BGRL representation learning (ablation) | Semi-supervised BGRL on MAG240M · MAG240M validation | classification accuracy: +73.89% | Helped |
| Use the same simple augmentations instead of complex adaptive augmentations (ablation) | BGRL · Coauthor CS, Coauthor Physics, Amazon Computers, Amazon Photos | classification accuracy: 93.31 ± 0.13 -> 93.34 ± 0.13; 95.73 ± 0.05 -> 95.62 ± 0.09; 90.34 ± 0.19 -> 90.39 ± 0.22; 93.17 ± 0.30 -> 93.15 ± 0.37 | No clear effect |
Closest papers by meaning
- OpenGSL: A Comprehensive Benchmark for Graph Structure Learning (NeurIPS 2023 · similarity 0.89)
- Learning the Latent Structure: A Feature-Centric Approach to Graph Data Augmentation (AAAI 2026 · similarity 0.88)
- Bootstrapping Heterogeneous Graph Representation Learning via Large Language Models: A Generalized Approach (AAAI 2025 · similarity 0.87)
- Exploitation of a Latent Mechanism in Graph Contrastive Learning: Representation Scattering (NeurIPS 2024 · similarity 0.87)
- Multiplex Graph Representation Learning with Homophily and Consistency (AAAI 2025 · similarity 0.87)
- Graph Classification via Reference Distribution Learning: Theory and Practice (NeurIPS 2024 · similarity 0.86)
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.