Summary of Catgnn: Cost-efficient and Scalable Distributed Training For Graph Neural Networks, by Xin Huang et al.
CATGNN: Cost-Efficient and Scalable Distributed Training for Graph Neural Networks
by Xin Huang, Weipeng Zhuo, Minh Phu Vuong, Shiju Li, Jongryool Kim, Bradley Rees, Chul-Ho Lee
First submitted to arxiv on: 2 Apr 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Distributed, Parallel, and Cluster Computing (cs.DC)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes a novel distributed Graph Neural Network (GNN) training system called CATGNN, designed to efficiently train GNNs on massive real-world graphs that exceed the memory capacity of commodity workstations. Unlike existing systems, which require loading entire graphs into memory for partitioning, CATGNN takes a stream of edges as input, reducing memory requirements. The system includes a novel streaming partitioning algorithm called SPRING, which significantly outperforms state-of-the-art algorithms and enables training on the largest publicly available dataset without requiring increased memory resources. Experimental results demonstrate the effectiveness and scalability of CATGNN with SPRING on 16 open datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about finding a way to train Graph Neural Networks (GNNs) on really big graphs that are too big for regular computers to handle. Right now, it’s hard to do this because we need to load the whole graph into memory, which takes up too much space. The authors of this paper came up with an idea called CATGNN that can take in edges one by one instead of loading the whole graph at once. This makes it possible to train GNNs on these huge graphs without needing super powerful computers. They also developed a new way to split up the graph, called SPRING, which is much better than what’s currently available. |
Keywords
* Artificial intelligence * Gnn * Graph neural network
