Exploring Zero-Shot Foundation Models for Multivariate Time Series Anomaly Detection
Martin Uray, Saverio Messineo, Roland Kwitt, Stefan Huber
Why It Matters
What makes this one worth your time
Understanding the limitations of zero-shot models in MTSAD can guide future research towards more effective anomaly detection techniques in critical domains.
The study reveals that zero-shot foundation models struggle with multivariate time series anomaly detection.
Summary
The paper investigates the application of a univariate forecasting foundation model, TimesFM, for multivariate time series anomaly detection (MTSAD) in industrial settings, evaluating two strategies but ultimately finding them ineffective compared to established baselines.
Key contributions
- Evaluation of zero-shot foundation models for MTSAD on the Secure Water Treatment benchmark.
- Comparison of two strategies for applying a univariate forecasting model to multivariate anomaly detection.
Notable insights
- Foundation models may excel at capturing temporal dynamics, leading to low error rates even in anomalous conditions, which complicates detection.
- Error peaks at anomaly boundaries suggest potential for using foundation models in change-point detection rather than direct anomaly detection.
Possible limitations
- The proposed setups were not competitive with established baselines, indicating a significant gap in performance.
- Not stated in the abstract.
Abstract
arXiv:2607.12454v1 Announce Type: new Abstract: Multivariate Time Series Anomaly Detection (MTSAD) is essential for reliability and safety in domains such as industrial process monitoring and financial risk management, yet conventional approaches rely on application-specific models that are costly to train and hard to scale. Foundation Models (FMs), pre-trained on broad data with strong zero-shot generalization, have recently become available for univariate time series forecasting, raising the question of whether they can address MTSAD without task-specific training. We investigate the zero-shot application of a univariate forecasting FM, TimesFM, to industrial MTSAD on the Secure Water Treatment (SWaT) benchmark, evaluating two strategies: treating the FM as a per-feature forecaster with thresholded prediction errors, and as an embedder whose intermediate representations feed standard outlier detectors. Neither of our proposed setups is competitive with established baselines; embeddings reveal only partial separation between normal and anomalous segments, insufficient for reliable detection. The cause is that the FM is too effective at capturing temporal dynamics, yielding low error even within fully anomalous windows, so persistent anomalies become indistinguishable from normal behavior. However, these observations yield valuable insights: the error peaks at anomaly boundaries, indicating FMs reliably detect distribution changes. We conclude that the proposed naive zero-shot FMs are unsuitable for MTSAD but promising for change-point detection.