Reproduce “CrossAD: Time Series Anomaly Detection with Cross-scale Associations and Cross-window Modeling” 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

Beibu Li, Qichao Shentu, Yang Shu, Hui Zhang, Ming Li, Ning Jin, Bin Yang, Chenjuan Guo

CrossAD introduces an unsupervised time series anomaly detection framework that models cross-scale associations and integrates cross-window global context. The system achieves state-of-the-art performance on seven real-world datasets, outperforming 18 baselines on VUS-PR and VUS-ROC metrics.

Key insights

  • Anomalies disrupt the predictable associations between coarse-grained and fine-grained representations of a time series, and explicitly modeling these cross-scale disruptions can improve detection.
  • Overcoming the limitations of fixed sliding windows by integrating a global multi-scale context allows for a more comprehensive understanding of temporal dependencies and context-dependent anomalies.
  • Learned sub-series queries can effectively represent recurring patterns, which, when used to update a global context, enhance the model's ability to share knowledge across different time windows.

Summary by alphaXiv.

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