Spatio-Temporal Mixed Graph Neural Controlled Differential Equations with Adaptive Connection Sampling for Irregular Multivariate Time Series Anomaly Detection

Xudong Jia; Wei Peng; Chiran Shen; Baokang Zhao; Peng Xun

ICASSP 2025accepted0 citations

Spatio-Temporal Mixed Graph Neural Controlled Differential Equations with Adaptive Connection Sampling for Irregular Multivariate Time Series Anomaly Detection

Xudong Jia, Wei Peng, Chiran Shen, Baokang Zhao, Peng Xun

Abstract

Multivariate time series data often demonstrate sparse and irregular characteristics in real-world signal processing applications, making anomaly detection challenging. This paper introduces STMG-AD, a spatio-temporal mixed graph neural controlled differential equation method with adaptive connection sampling, designed specifically for anomaly detection in irregular multivariate time series. By integrating causal graphs with graph attention networks and employing adaptive connection sampling coupled with Monte Carlo dropout, STMG-AD can enhance the robustness and accuracy of anomaly detection. Experiments on various real-world datasets demonstrate its superiority over existing methods in detecting anomalies in irregular multivariate time series.

BibTeX

@inproceedings{icassp2025_spatiotemporalmi,
  title = {Spatio-Temporal Mixed Graph Neural Controlled Differential Equations with Adaptive Connection Sampling for Irregular Multivariate Time Series Anomaly Detection},
  author = {Xudong Jia and Wei Peng and Chiran Shen and Baokang Zhao and Peng Xun},
  booktitle = {ICASSP 2025},
  year = {2025}
}