Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DACR: Distribution-Augmented Contrastive Reconstruction for Time-Series Anomaly Detection

Lixu Wang, Shichao Xu, Xinyu Du, Qi Zhu

TL;DR

Time-series anomaly detection is challenged by multi-distribution normal data and diverse anomaly types. The authors propose Distribution-Augmented Contrastive Reconstruction (DACR), a three-stage framework that combines VAE-based distribution augmentation, contrastive learning for intrinsic semantics of univariate features, and a transformer-based reconstruction that models inter-feature dependencies. DACR addresses practical scenarios with distributional shifts, leverages self-supervised learning with attention-based reconstruction, and demonstrates state-of-the-art performance across nine benchmark datasets. This approach enhances robustness to distributional changes and improves anomaly sensitivity, with tangible implications for real-world monitoring systems.

Abstract

Anomaly detection in time-series data is crucial for identifying faults, failures, threats, and outliers across a range of applications. Recently, deep learning techniques have been applied to this topic, but they often struggle in real-world scenarios that are complex and highly dynamic, e.g., the normal data may consist of multiple distributions, and various types of anomalies may differ from the normal data to different degrees. In this work, to tackle these challenges, we propose Distribution-Augmented Contrastive Reconstruction (DACR). DACR generates extra data disjoint from the normal data distribution to compress the normal data's representation space, and enhances the feature extractor through contrastive learning to better capture the intrinsic semantics from time-series data. Furthermore, DACR employs an attention mechanism to model the semantic dependencies among multivariate time-series features, thereby achieving more robust reconstruction for anomaly detection. Extensive experiments conducted on nine benchmark datasets in various anomaly detection scenarios demonstrate the effectiveness of DACR in achieving new state-of-the-art time-series anomaly detection.

DACR: Distribution-Augmented Contrastive Reconstruction for Time-Series Anomaly Detection

TL;DR

Time-series anomaly detection is challenged by multi-distribution normal data and diverse anomaly types. The authors propose Distribution-Augmented Contrastive Reconstruction (DACR), a three-stage framework that combines VAE-based distribution augmentation, contrastive learning for intrinsic semantics of univariate features, and a transformer-based reconstruction that models inter-feature dependencies. DACR addresses practical scenarios with distributional shifts, leverages self-supervised learning with attention-based reconstruction, and demonstrates state-of-the-art performance across nine benchmark datasets. This approach enhances robustness to distributional changes and improves anomaly sensitivity, with tangible implications for real-world monitoring systems.

Abstract

Anomaly detection in time-series data is crucial for identifying faults, failures, threats, and outliers across a range of applications. Recently, deep learning techniques have been applied to this topic, but they often struggle in real-world scenarios that are complex and highly dynamic, e.g., the normal data may consist of multiple distributions, and various types of anomalies may differ from the normal data to different degrees. In this work, to tackle these challenges, we propose Distribution-Augmented Contrastive Reconstruction (DACR). DACR generates extra data disjoint from the normal data distribution to compress the normal data's representation space, and enhances the feature extractor through contrastive learning to better capture the intrinsic semantics from time-series data. Furthermore, DACR employs an attention mechanism to model the semantic dependencies among multivariate time-series features, thereby achieving more robust reconstruction for anomaly detection. Extensive experiments conducted on nine benchmark datasets in various anomaly detection scenarios demonstrate the effectiveness of DACR in achieving new state-of-the-art time-series anomaly detection.
Paper Structure (9 sections, 8 equations, 3 figures, 2 tables)

This paper contains 9 sections, 8 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Distribution-Augmented Contrastive Learning
  4. Intrinsic Dependency-Aware Reconstruction
  5. Experimental Results
  6. Experimental Settings
  7. Performance Comparison
  8. Ablation Study and Sensitivity Analysis
  9. Conclusion

Figures (3)

  • Figure 1: Overall workflow of Distribution-Augmented Contrastive Reconstruction (DACR) for time-series anomaly detection.
  • Figure 2: Performance comparison between DACR and baselines in the settings of both EAD (n=$N_C-$1) and IAD. AUC$\pm$standard deviation is used to evaluate the performance. DACR significantly outperforms the second-best by 2.8-8.2% on EAD, and 0.4-3.3% on IAD.
  • Figure 3: Performance comparison in EAD with different normal class numbers on the SAD dataset.