Reproduce “Graph Anomaly Detection in Time Series: 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
Thi Kieu Khanh Ho, Ali Karami, Narges Armanfard
A comprehensive survey consolidates research on Graph Anomaly Detection in Time Series (G-TSAD), presenting the first dedicated review of methods that leverage graph representations to capture both temporal and inter-variable dependencies. It systematically categorizes existing techniques based on their loss functions, critically analyzes current evaluation practices, and outlines key challenges and future directions for the field.
Key insights
- Graph representations are a highly effective paradigm for Time-Series Anomaly Detection, adept at modeling both intra-variable temporal and inter-variable relational dependencies.
- Current G-TSAD methods typically target specific anomaly types, with no existing study simultaneously detecting all five identified graph-based anomalies, and anomalous Sim{·, ·} (global relations between observations) remaining largely underexplored.
- Widely used benchmark datasets and evaluation metrics in TSAD often suffer from critical flaws like mislabeled ground truth, trivial anomalies, and overestimation due to methods like "point adjustment".
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.