Multivariate Time-series Anomaly Detection via Graph Attention Network

By Hang Zhao, Yujing Wang, Juanyong Duan, Congrui Huang, Defu Cao, Yunhai Tong, Bixiong Xu, Jing Bai, Jie Tong, Qi Zhang. Industrial Conference on Data Mining, 2020

722 citations recorded by the source corpus.

Anomaly detection on multivariate time-series is of great importance in both data mining research and industrial applications. Recent approaches have achieved significant progress in this topic, but there is remaining limitations. One major limitation is that they do not capture the relationships between different time-series explicitly, resulting in inevitable false alarms. In this paper, we propose a novel self-supervised framework for multivariate time-series anomaly detection to address this issue. Our framework considers each univariate time-series as an individual feature and includes two graph attention layers in parallel to learn the complex dependencies of multivariate time-series in both temporal and feature dimensions. In addition, our approach jointly optimizes a forecasting-based model and a reconstruction-based model, obtaining better time-series representations through a combination of single-timestamp prediction and reconstruction of the entire time-series. We demonstrate the efficacy of our model through extensive experiments. The proposed method outperforms other state-of-the-art models on three real-world datasets. Further analysis shows that our method has good interpretability and is useful for anomaly diagnosis.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/8611d3666f73011662671f48591c076959ee7383

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

A Survey on Graph Neural Networks for Time Series: Forecasting, Classification, Imputation, and Anomaly Detection (2023, methodology): “Ingeneral, thespatial-temporalGNNandthepredictoraretrainedonnor-mal,non-anomalousdatausingeitherforecasting[36],[56]or reconstruction[35],[57]optimizationapproaches,withtheaim ofminimizingthediscrepancybetweenthenormalinputandthe forecast(orreconstructed)series.”
Anomaly Detection in Time Series: A Comprehensive Evaluation (2022, methodology): “From the 71 implemented algorithms, five algorithms could not process a single dataset due to hitting memory or time limits [43, 52, 86, 159, 161].”
Anomaly Transformer: Time Series Anomaly Detection with Association Discrepancy (2021, methodology): “Recently, graph neural network (GNN) has been applied to learn the dynamic graph among multiple variables in multivariate time series (Zhao et al., 2020; Deng & Hooi, 2021).”
Deep Learning for Time Series Anomaly Detection: A Survey (2022, neutral): “In [195], two parallel graph attention (GAT) layers are proposed for self-supervised multivariate time series anomaly detection, which will extract correlations between different time series (intermetric dependencies) and learn the relationships between timestamps (temporal dependencies).”
Deep Learning for Anomaly Detection in Time-Series Data: Review, Analysis, and Guidelines (2021, methodology): “Recently, several approaches [73], [108], [109] that applied an attention mechanism to GNN have been proposed to improve performance for identifying root causes.”
Learning Graph Structures With Transformer for Multivariate Time-Series Anomaly Detection in IoT (2021, methodology): “(4) Though GDN [4] is also a graph learning-based anomaly detection approach, it adopts the top-K nearest connection strategy to model the topological graph structure among sensors, which have certain limitations as we discussed in Section I. MTAD-GAT [17] directly utilizes the initial graph structu”
DCdetector: Dual Attention Contrastive Representation Learning for Time Series Anomaly Detection (2023, background): “Most of the supervised and semi-supervised methods [14, 18, 51, 53, 87, 90] can not handle the challenge of limited labeled data, especially the anomalies are dynamic and new anomalies never observed before may occur.”