Eliminating Data Processing Bottlenecks in GNN Training over Large Graphs via Two-level Feature Compression
Summary: F2CGT removes sampling and feature-loading bottlenecks via two-level hybrid feature compression (per-node compressor choice) with provable convergence and negligible accuracy loss (up to 128× compression). A cost-model GPU cache and memory partitioning yields 1.2–2.6× single-node and 3.6–71× distributed training speedups. (summarized by gpt-5-mini on Feb 09 2026)
Incoming Non-self Citations Over Time
Authors
- 1. Yuxin Ma
- 2. Ping Gong
- 3. Tianming Wu
- 4. Jiawei Yi
- 5. Chengru Yang
- 6. Cheng Li
- 7. Qirong Peng
- 8. Guiming Xie
- 9. Yongcheng Bao
- 10. Haifeng Liu
- 11. Yinlong Xu
Incoming Citations (Sorted by Pagerank)
Showing 2 of 2 citing papers.
| Rank | Citing Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 7,607 | Systems for Scalable Graph Analytics and Machine Learning: Trends and Methods | 2025 | VLDB | 4.6967024e-05 |
| 10,735 | Faster Convergence in Mini-batch Graph Neural Networks Training with Pseudo Full Neighborhood Compensation | 2025 | VLDB | 4.1945683e-05 |
Previous
Page 1 / 1
Next
Outgoing Citations (Sorted by Pagerank)
Showing 1 of 1 cited papers.
Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.
| Rank | Cited Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 2,422 | DUCATI: A Dual-Cache Training System for Graph Neural Networks on Giant Graphs with the GPU | 2023 | SIGMOD | 8.8499665e-05 |
Previous
Page 1 / 1
Next