PRISM: A Training System to Unlock the Potential of Temporal Graph Learning Through Staleness Avoidance

Summary: PRISM eliminates temporal-memory staleness in batched M-TGNN training via multi-versioned memory and relaxed “lazy freshness,” preserving GPU parallelism. It substantially improves accuracy over stale and strict staleness-free systems while maintaining competitive training time. (summarized by gpt-5.6-luna on Aug 17 2026)

Paper ID: hf40fd929655b5b82
Venue: VLDB
Year: 2026
Pagerank: 4.9793485e-05
Overall Rank: 10,779 | 27.53%
DOI: 10.14778/3819518.3819544

Incoming Non-self Citations Over Time

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

Authors

1. Md Ashraful Islam (University of Massachusetts Amherst)
2. Hojae Son (University of Massachusetts Amherst)
3. Suhaas Kiran Doddagaddavalli Gangadharaiah (University of Massachusetts Amherst)
4. Marco Serafini (University of Massachusetts Amherst)

BibTeX Citation

@article{islam_vldb26,
        title = {{PRISM: A Training System to Unlock the Potential of Temporal Graph Learning Through Staleness Avoidance}},
        author = {Islam, Md Ashraful and Son, Hojae and Gangadharaiah, Suhaas Kiran Doddagaddavalli and Serafini, Marco},
        journal = {PVLDB},
        series = {{VLDB} '26},
        volume = {19},
        number = {9},
        pages = {2196--2209},
        doi = {10.14778/3819518.3819544},
        url = {https://doi.org/10.14778/3819518.3819544},
        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 9 of 9 cited papers.

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

Rank	Cited Paper	Year	Venue	Pagerank
881	APAN: Asynchronous Propagation Attention Network for Real-time Temporal Graph Embedding	2021	SIGMOD	0.00013278822
1,364	TGL: A General Framework for Temporal GNN Training on Billion-Scale Graphs	2022	VLDB	0.00010914152
3,017	Zebra: When Temporal Graph Neural Networks Meet Temporal Personalized PageRank	2023	VLDB	7.7471344e-05
3,399	Orca: Scalable Temporal Graph Neural Network Training with Theoretical Guarantees	2023	SIGMOD	7.3349076e-05
4,502	NeutronStream: A Dynamic GNN Training Framework with Sliding Window for Graph Streams	2024	VLDB	6.5740722e-05
4,764	ETC: Efficient Training of Temporal Graph Neural Networks over Large-scale Dynamic Graphs	2024	VLDB	6.4276208e-05
5,310	DynaHB: A Communication-Avoiding Asynchronous Distributed Framework with Hybrid Batches for Dynamic GNN Training	2024	VLDB	6.1857447e-05
9,232	PipeTGL: (Near) Zero Bubble Memory-based Temporal Graph Neural Network Training via Pipeline Optimization	2025	VLDB	5.2056825e-05
9,234	Effective and Efficient Distributed Temporal Graph Learning through Hotspot Memory Sharing	2025	VLDB	5.2056825e-05

Semantically Similar Papers

#	Overall Rank	Paper	Year	Venue
1	10,706	Efficient GNN Training on Giant Graphs with Collective Batching and Scheduling	2026	VLDB
2	8,677	PrismX: A Single-Machine System for Querying Big Graphs	2024	VLDB
3	10,737	FlareDTDG: Harnessing Temporal Recency for Scalable Discrete-Time Dynamic Graph Training	2026	VLDB
4	9,234	Effective and Efficient Distributed Temporal Graph Learning through Hotspot Memory Sharing	2025	VLDB
5	9,232	PipeTGL: (Near) Zero Bubble Memory-based Temporal Graph Neural Network Training via Pipeline Optimization	2025	VLDB
6	6,826	SIMPLE: Efficient Temporal Graph Neural Network Training at Scale with Dynamic Data Placement	2024	SIGMOD
7	10,535	SWIFT: Enabling Large-Scale Temporal Graph Learning on a Single Machine	2026	SIGMOD
8	11,459	Towards Ideal Temporal Graph Neural Networks: Evaluations and Conclusions after 10,000 GPU Hours	2025	VLDB
9	5,339	FreshGNN: Reducing Memory Access via Stable Historical Embeddings for Graph Neural Network Training	2024	VLDB
10	1,364	TGL: A General Framework for Temporal GNN Training on Billion-Scale Graphs	2022	VLDB