AG 1, AG 2, AG 3, INET, AG 4, AG 5, D6, SWS, RG1, MMCI   Info

AG Audience

English

30 Minutes

Time series foundational models (TSFM) have gained significant attention for their promise of state-of-the-art performance in time series forecasting across various applications. However, their use in anomaly detection and prediction has not been thoroughly explored, and growing concerns persist regarding their black-box nature, lack of interpretability, and practical applicability. In this talk, I will present findings that critically assess the effectiveness of TSFM for anomaly detection and prediction tasks. Here I will highlight the key limitations of TSFM through systematic analysis across multiple datasets, including those with no clear patterns, trends, or seasonality. We found that, while they can be extended to anomaly detection and prediction, traditional statistical and deep learning models often perform equally well or outperform TSFM. Furthermore, TSFMs tend to require high computational resources without effectively capturing sequential dependencies, and their performance is not significantly better in few-shot or zero-shot scenarios. Through this analysis we shed light on the practical challenges of applying TSFM to anomaly detection and prediction and offer insights into where traditional models may still hold an edge. Link to the paper  - Paper link

Ina Geisler

+49 681 9325 1802

--email hidden

Zoom

passcode not visible

logged in users only