![[2602.19068] TimeRadar: A Domain-Rotatable Foundation Model for Time Series Anomaly Detection](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19068] TimeRadar: A Domain-Rotatable Foundation Model for Time Series Anomaly Detection
Summary
TimeRadar introduces a novel approach to time series anomaly detection using a domain-rotatable foundation model that enhances the differentiation of normal and abnormal patterns across datasets.
Why It Matters
This research addresses a critical gap in time series anomaly detection by proposing a model that adapts to various datasets, improving the accuracy of identifying irregular patterns. As businesses increasingly rely on data-driven decisions, effective anomaly detection becomes essential for operational integrity and risk management.
Key Takeaways
- TimeRadar utilizes a fractional time-frequency domain for anomaly detection.
- The model can adaptively differentiate between normal and abnormal signals.
- Introduces FTFRecon for optimal data reconstruction in diverse datasets.
- Includes Contextual Deviation Learning to capture local anomalies.
- Addresses limitations of existing time series foundation models.
Computer Science > Machine Learning arXiv:2602.19068 (cs) [Submitted on 22 Feb 2026] Title:TimeRadar: A Domain-Rotatable Foundation Model for Time Series Anomaly Detection Authors:Hui He, Hezhe Qiao, Yutong Chen, Kun Yi, Guansong Pang View a PDF of the paper titled TimeRadar: A Domain-Rotatable Foundation Model for Time Series Anomaly Detection, by Hui He and 3 other authors View PDF HTML (experimental) Abstract:Current time series foundation models (TSFMs) primarily focus on learning prevalent and regular patterns within a predefined time or frequency domain to enable supervised downstream tasks (e.g., forecasting). Consequently, they are often ineffective for inherently unsupervised downstream tasks-such as time series anomaly detection (TSAD), which aims to identify rare, irregular patterns. This limitation arises because such abnormal patterns can closely resemble the regular patterns when presented in the same time/frequency domain. To address this issue, we introduce TimeRadar, an innovative TSFM built in a fractional time-frequency domain to support generalist TSAD across diverse unseen datasets. Our key insight is that rotating a time series into a data-dependent fractional time-frequency representation can adaptively differentiate the normal and abnormal signals across different datasets. To this end, a novel component, namely Fractionally modulated Time-Frequency Reconstruction (FTFRecon), is proposed in TimeRadar to leverage a learnable fractional order to rotat...
