Eliminating Data Processing Bottlenecks in GNN Training over Large Graphs via Two-level Feature Compression

Summary: F²CGT removes GNN training’s sampling/feature-loading bottleneck using node-differentiated, two-level feature compression with convergence guarantees. A cost-model-driven GPU cache co-design reaches 128× compression and 1.23–71.46× speedups with marginal accuracy loss. (summarized by gpt-5.6-luna on Jul 24 2026)

Paper ID: 13694
Venue: VLDB
Year: 2024
Pagerank: 5.8139657e-05
Overall Rank: 6,652 | 54.37%
DOI: 10.14778/3681954.3681968

Incoming Non-self Citations Over Time

Authors

1. Yuxin Ma (University of Science and Technology Beijing)
2. Ping Gong (University of Science and Technology Beijing)
3. Tianming Wu (University of Science and Technology Beijing)
4. Jiawei Yi (University of Science and Technology Beijing)
5. Chengru Yang (University of Science and Technology Beijing)
6. Cheng Li (Hefei Comprehensive National Science Center; University of Science and Technology Beijing)
7. Qirong Peng (OPPO)
8. Guiming Xie (OPPO)
9. Yongcheng Bao (OPPO)
10. Haifeng Liu (OPPO)
11. Yinlong Xu (University of Science and Technology Beijing)

BibTeX Citation

@article{ma_vldb24,
        title = {{Eliminating Data Processing Bottlenecks in GNN Training over Large Graphs via Two-level Feature Compression}},
        author = {Ma, Yuxin and Gong, Ping and Wu, Tianming and Yi, Jiawei and Yang, Chengru and Li, Cheng and Peng, Qirong and Xie, Guiming and Bao, Yongcheng and Liu, Haifeng and Xu, Yinlong},
        journal = {PVLDB},
        series = {{VLDB} '24},
        volume = {17},
        number = {11},
        pages = {2854--2866},
        doi = {10.14778/3681954.3681968},
        url = {https://doi.org/10.14778/3681954.3681968},
        year = {2024}
}

Incoming Citations (Sorted by Pagerank)

Showing 3 of 3 citing papers.

Rank	Citing Paper	Year	Venue	Pagerank
7,601	Systems for Scalable Graph Analytics and Machine Learning: Trends and Methods	2025	VLDB	5.5866563e-05
10,241	FastGNAS: Accelerating and Scaling Graph Neural Architecture Search on Multi-GPUs via Ring-Based Model Migration	2026	SIGMOD	5.093636e-05
10,975	Faster Convergence in Mini-batch Graph Neural Networks Training with Pseudo Full Neighborhood Compensation	2025	VLDB	5.093636e-05

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
2,668	DUCATI: A Dual-Cache Training System for Graph Neural Networks on Giant Graphs with the GPU	2023	SIGMOD	8.2750247e-05

Semantically Similar Papers

#	Overall Rank	Paper	Year	Venue
1	1,863	ByteGNN: Efficient Graph Neural Network Training at Large Scale	2022	VLDB
2	7,353	XGNN: Boosting Multi-GPU GNN Training via Global GNN Memory Store	2024	VLDB
3	10,975	Faster Convergence in Mini-batch Graph Neural Networks Training with Pseudo Full Neighborhood Compensation	2025	VLDB
4	1,234	Large Graph Convolutional Network Training with GPU-Oriented Data Communication Architecture	2021	VLDB
5	5,316	Comprehensive Evaluation of GNN Training Systems: A Data Management Perspective	2024	VLDB
6	6,630	Efficient Training of Graph Neural Networks on Large Graphs	2024	VLDB
7	2,640	Scalable and Efficient Full-Graph GNN Training for Large Graphs	2023	SIGMOD
8	5,597	Accelerating Sampling and Aggregation Operations in GNN Frameworks with GPU Initiated Direct Storage Accesses	2024	VLDB
9	10,913	Inference-friendly Graph Compression for Graph Neural Networks	2025	VLDB
10	5,243	FreshGNN: Reducing Memory Access via Stable Historical Embeddings for Graph Neural Network Training	2024	VLDB