FAST framework accelerates Temporal Graph Neural Network training

Researchers have developed FAST, a new framework designed to optimize the training of Temporal Graph Neural Networks (TGNNs). TGNNs are crucial for analyzing dynamic graphs in areas like recommendations and social network analysis, but their training is hindered by memory I/O, computation, and sampling bottlenecks. FAST addresses these issues by holistically optimizing these stages together, introducing SlimCache for efficient memory management under GPU constraints and employing specialized operators for sparse temporal subgraphs. The framework also utilizes a topology-aware sampling strategy to improve CPU cache locality. Experiments show FAST achieves significant speedups, averaging 2.1x and up to 4.7x, without compromising model accuracy. AI

IMPACT This framework could significantly speed up training for graph-based AI models, enabling larger and more complex analyses in recommendation systems and social network analysis.

RANK_REASON The cluster describes a research paper detailing a new framework for optimizing TGNN training.

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• central processing unit
• DagsHub
• Gotit.pub
• graphics processing unit
• Hugging Face
• ScienceCast
• SlimCache
• Temporal Graph Neural Networks
• TGNNs

• paper
• infra

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →