YOSO: You-Only-Sample-Once via Compressed Sensing for Graph Neural Network Training
Keywords: Sampling, Graph Neural Network, Compressed Sensing
TL;DR: Use compressed sensing technique to develop a sampling method that can reduce sampling and therefore, total training time in GNN.
Abstract: Graph Neural Networks (GNNs) have become essential tools for analyzing structured data across various domains. In GNNs, sampling is critical for reducing training latency by limiting the number of nodes processed during training, especially for large-scale applications. However, as the demand for better prediction performance increases, existing sampling algorithms become more complex, introducing significant overhead in the training process. To address this issue, we introduce YOSO (You-Only-Sample-Once), an algorithm designed to achieve highly efficient training while preserving prediction accuracy in downstream tasks. YOSO proposes a compressed sensing-based sampling and reconstruction framework, where nodes are sampled once at the input layer, followed by a lossless reconstruction at the output layer during each epoch. This approach not only avoids costly computations, such as orthonormal basis, but also guarantees high-probability accuracy retention, equivalent to full node participation. Experimental results on both node classification and link prediction tasks demonstrate the effectiveness and efficiency of YOSO, reducing GNN training by an average of around 75% compared to state-of-the-art methods, while maintaining accuracy on par with top-performing baselines.
Primary Area: learning on graphs and other geometries & topologies
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4725
Loading