Reproduce “Explainable Spatio-Temporal Graph Neural Networks” 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

Jiabin Tang, Lianghao Xia, Chao Huang

University of Hong Kong researchers developed STExplainer, a framework that integrates explainability directly into Spatio-Temporal Graph Neural Networks, achieving state-of-the-art predictive accuracy while providing faithful explanations for spatio-temporal predictions. The framework identifies influential spatial and temporal subgraphs underlying predictions, demonstrating robustness to data missingness and sparsity on traffic and crime datasets.

Key insights

  • Intrinsic explainability, achieved through a structure-distilled Graph Information Bottleneck (GIB), can simultaneously enhance prediction accuracy and provide faithful explanations in Spatio-Temporal Graph Neural Networks (STGNNs).
  • The GIB principle effectively filters out task-irrelevant information and noise, contributing to STExplainer's robustness against data missingness and sparsity in real-world scenarios.
  • Meaningful spatio-temporal explainability for STGNNs can be defined by identifying influential spatial and temporal subgraphs and quantitatively evaluated using adapted Sparsity and Fidelity metrics.

Summary by alphaXiv.

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