DUAL-TASK VAE FOR NODE-LEVEL DATA AUGMENTATION
Keywords: VAE, GNNs, Graph Data Augmentation, Node Classification, Dual-task Training, Loss Weight Adjustment
TL;DR: Introducing valuable strategies for enhancing graph data augmentation, enabling simpler models to achieve better performance.
Abstract: Graph Neural Networks (GNNs) have shown great promise in processing graph-structured data, but they often require large amounts of labeled data and are sensitive to noise. In this paper, we propose a novel node-level data augmentation approach that leverages a Variational Autoencoder (VAE) within a dual-task learning framework to address these challenges. Our method utilizes the VAE to generate enriched node representations that capture both structural and feature-related information, which are then combined with the original node features for classification by a Graph Attention Network (GAT). Experiments conducted on the Cora, Citeseer, and Pubmed datasets show that our approach outperforms baseline models, achieving up to 7.3\% higher accuracy in Pubmed, and surpassing recent state-of-the-art data augmentation techniques. This work highlights the effectiveness of dual-task learning for robust feature enhancement and advances data augmentation strategies in GNNs.
Supplementary Material: zip
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.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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: 1503
Loading