FlareDTDG: Harnessing Temporal Recency for Scalable Discrete-Time Dynamic Graph Training

Summary: FlareDTDG: distributed DTDG training that exploits temporal recency via hybrid batching with temporal decay—full-batch on recent snapshots, coarse sampling on older ones. Adds shrinking-based graph reconstruction and adaptive comm/comp overlap for 1.4–2.5x speedups, 10–85% less GPU memory, near-lossless accuracy. (summarized by gpt-5.4-mini on May 27 2026)

Paper ID: 14492
Venue: VLDB
Year: 2026
Pagerank: 5.093636e-05
Overall Rank: 10,555 | 27.59%
DOI: 10.14778/3801059.3801073

Incoming Non-self Citations Over Time

No non-self incoming citations found for this paper in this database.

Authors

1. Wenjie Huang (Zhejiang University)
2. Rui Wang (Zhejiang University)
3. Jing Cao (Hangzhou City University; Zhejiang University)
4. Tongya Zheng (Hangzhou City University)
5. Xinyu Wang (Zhejiang University)
6. Mingli Song (Zhejiang University)
7. Sai Wu (Zhejiang University)
8. Chun Chen (Zhejiang University)

BibTeX Citation

@article{huang_vldb26,
        title = {{FlareDTDG: Harnessing Temporal Recency for Scalable Discrete-Time Dynamic Graph Training}},
        author = {Huang, Wenjie and Wang, Rui and Cao, Jing and Zheng, Tongya and Wang, Xinyu and Song, Mingli and Wu, Sai and Chen, Chun},
        journal = {PVLDB},
        series = {{VLDB} '26},
        volume = {19},
        number = {7},
        pages = {1614--1627},
        doi = {10.14778/3801059.3801073},
        url = {https://doi.org/10.14778/3801059.3801073},
        year = {2026}
}

Incoming Citations (Sorted by Pagerank)

Showing 0 of 0 citing papers.

Rank	Citing Paper	Year	Venue	Pagerank

Outgoing Citations (Sorted by Pagerank)

Showing 4 of 4 cited papers.

Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.

Rank	Cited Paper	Year	Venue	Pagerank
1,132	SANCUS: Staleness-Aware Communication-Avoiding Full-Graph Decentralized Training in Large-Scale Graph Neural Networks	2022	VLDB	0.00012041292
4,891	DGC: Training Dynamic Graphs with Spatio-Temporal Non-Uniformity using Graph Partitioning by Chunks	2023	SIGMOD	6.4581865e-05
5,760	DynaHB: A Communication-Avoiding Asynchronous Distributed Framework with Hybrid Batches for Dynamic GNN Training	2024	VLDB	6.0961929e-05
10,907	Effective and Efficient Distributed Temporal Graph Learning through Hotspot Memory Sharing	2025	VLDB	5.093636e-05

Semantically Similar Papers

#	Overall Rank	Paper	Year	Venue
1	10,780	SWASH: A Flexible Communication Framework with Sliding Window-Based Cache Sharing for Scalable DGNN Training	2025	SIGMOD
2	4,980	Decoupled Graph Neural Networks for Large Dynamic Graphs	2023	VLDB
3	11,110	Towards Ideal Temporal Graph Neural Networks: Evaluations and Conclusions after 10,000 GPU Hours	2025	VLDB
4	5,760	DynaHB: A Communication-Avoiding Asynchronous Distributed Framework with Hybrid Batches for Dynamic GNN Training	2024	VLDB
5	5,205	ETC: Efficient Training of Temporal Graph Neural Networks over Large-scale Dynamic Graphs	2024	VLDB
6	6,730	SIMPLE: Efficient Temporal Graph Neural Network Training at Scale with Dynamic Data Placement	2024	SIGMOD
7	1,409	TGL: A General Framework for Temporal GNN Training on Billion-Scale Graphs	2022	VLDB
8	4,891	DGC: Training Dynamic Graphs with Spatio-Temporal Non-Uniformity using Graph Partitioning by Chunks	2023	SIGMOD
9	10,907	Effective and Efficient Distributed Temporal Graph Learning through Hotspot Memory Sharing	2025	VLDB
10	10,620	Understanding Evolving Graph Structures for Large Discrete-Time Dynamic Graph Representation	2026	VLDB