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

arXiv cs.AI · 6 hours ago

VAN-AD is a new framework that adapts vision-based Masked Autoencoders to detect anomalies in time series data, addressing generalization problems in existing methods that require separate models per dataset. The approach was tested on nine real-world datasets and consistently outperformed state-of-the-art methods across multiple metrics. The framework enables more effective anomaly detection in IoT systems and other scenarios with limited training data.