Reproduce “Graph Data Augmentation for Graph Machine Learning: A Survey” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Tong Zhao, Wei Jin, Yozen Liu, Yingheng Wang, Gang Liu, Stephan Günnemann, Neil Shah, Meng Jiang

Researchers from Snap Inc., Michigan State University, University of Notre Dame, Cornell University, and Technical University of Munich present a comprehensive survey of Graph Data Augmentation (GDA) methods, systematizing the field through multi-dimensional taxonomies and detailing both rule-based and learned techniques. The work clarifies how GDA addresses graph machine learning challenges like data sparsity and noise, highlighting its crucial role in self-supervised learning and identifying key open problems for future research.

Key insights

  • GDA is essential for mitigating GML-specific challenges such as data sparsity, noise, and label scarcity, thereby enhancing GML model performance, generalization, and robustness.
  • Both simple rule-based corruptions and sophisticated learned augmentation strategies are widely adopted in various GML tasks and are foundational for self-supervised learning on graphs.
  • The field requires significant future work in developing domain adaptation methods, improving scalability for large-scale graphs, establishing comprehensive evaluation criteria, and building a rigorous theoretical foundation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.