VAN-AD: Visual Masked Autoencoder With Normalizing Flow For Time Series Anomaly Detection

Author

• Peng Yu Chen
• Shang Wang
• Xiaohou Shi
• Yuan Chang
• Yan Sun
• Sajal K. Das, Missouri University of Science and TechnologyFollow

Abstract

Time series anomaly detection (TSAD) is essential for reliable and secure IoT-enabled service systems. However, dataset-specific training limits the generalization of existing methods, especially when target-domain training data are scarce. Although foundation models offer a promising direction, current attempts based on large language models or large-scale time series pre-training still suffer from cross-modal gaps or in domain heterogeneity. In this paper, we investigate the applicability of large-scale pre-trained vision models to TSAD. Specifically, we adapt a visual Masked Autoencoder (MAE) pre-trained on ImageNet to the TSAD task. However, directly transferring MAE to TSAD introduces two key challenges: over generalization and limited local perception. To address these challenges, we propose VAN-AD, a framework that adapts an ImageNet-pre-trained visual MAE to TSAD, which combines adaptive reconstruction post-processing and flow-based density estimation. To alleviate the over-generalization issue, we design an Adaptive Distribution Modulation Module (ADMM) as a training-free post-processing operation, which recalibrates the MAE reconstruction using the local statistics of the input window to amplify discrepancies caused by abnormal patterns. To overcome the limitation of local perception, we further develop a Normalizing Flow Module (NFM), which combines MAE with normalizing flow to estimate the probability density of the current window under the global distribution. Extensive experiments on nine real-world datasets demonstrate that VAN-AD consistently outperforms existing state-of-the-art methods across multiple evaluation metrics.

Recommended Citation

P. Y. Chen et al., "VAN-AD: Visual Masked Autoencoder With Normalizing Flow For Time Series Anomaly Detection," IEEE Transactions on Dependable and Secure Computing, Institute of Electrical and Electronics Engineers, Jan 2026.

The definitive version is available at https://doi.org/10.1109/TDSC.2026.3732446

Department(s)

Computer Science

Publication Status

Early Access

Keywords and Phrases

IoT service systems; Masked autoencoders; Normalizing flow; Time series anomaly detection; Vision foundation models

International Standard Serial Number (ISSN)

1941-0018; 1545-5971

Document Type

Article - Journal

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2026 Institute of Electrical and Electronics Engineers, All rights reserved.

Publication Date

01 Jan 2026