mlsanityData to 7 Oct 2026

Performance of Zero-Shot Time Series Foundation Models on Cloud Data

arXiv 2502.12944 ↗First posted 18 Feb 2025cs.LGcited by 10

Not on the map. Why?

Abstract (arXiv)

Time series foundation models (FMs) have emerged as a popular paradigm for zero-shot multi-domain forecasting. FMs are trained on numerous diverse datasets and claim to be effective forecasters across multiple different time series domains, including cloud data. In this work we investigate this claim, exploring the effectiveness of FMs on cloud data. We demonstrate that many well-known FMs fail to generate meaningful or accurate zero-shot forecasts in this setting. We support this claim empirically, showing that FMs are outperformed consistently by simple linear baselines. We also illustrate a number of interesting pathologies, including instances where FMs suddenly output seemingly erratic, random-looking forecasts. Our results suggest a widespread failure of FMs to model cloud data.

Key facts. Cited by 10 papers in the map (data through 7 Oct 2026). Builds on 17 of its 26 references that are in the map. First posted on arXiv 18 Feb 2025.

Builds on

17 of 26 references.

References · 17

  1. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,339
  2. 2022A Time Series is Worth 64 Words: Long-term Forecasting with Transformerscited by 1,981
  3. 2021Domain Generalization: A Surveycited by 760
  4. 2024Chronos: Learning the Language of Time Seriescited by 712
  5. 2023A decoder-only foundation model for time-series forecastingcited by 650
  6. 2024Unified Training of Universal Time Series Forecasting Transformerscited by 588
  7. 2024Foundation Models for Time Series Analysis: A Tutorial and Surveycited by 234
  8. 2024GIFT-Eval: A Benchmark For General Time Series Forecasting Model Evaluationcited by 188
  9. 2024Tiny Time Mixers (TTMs): Fast Pre-trained Models for Enhanced Zero/Few-Shot Forecasting of Multivariate Time Seriescited by 120
  10. 2024VisionTS: Visual Masked Autoencoders Are Free-Lunch Zero-Shot Time Series Forecasterscited by 87
  11. 2024TSFM-Bench: A Comprehensive and Unified Benchmark of Foundation Models for Time Series Forecastingcited by 55
  12. 2024An Analysis of Linear Time Series Forecasting Modelscited by 40
5 more
  1. 2024Empowering Time Series Analysis with Foundation Models: A Comprehensive Surveycited by 23
  2. 2024Mamba4Cast: Efficient Zero-Shot Time Series Forecasting with State Space Modelscited by 17
  3. 2023How Does It Function? Characterizing Long-term Trends in Production Serverless Workloadscited by 10
  4. 2024DAM: Towards A Foundation Model for Time Series Forecastingcited by 10
  5. 2024Serverless Cold Starts and Where to Find Themcited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 6; 2026: 4.

10 papers.

  1. 2025This Time is Different: An Observability Perspective on Time Series Foundation Modelscited by 57
  2. 2025True Zero-Shot Inference of Dynamical Systems Preserving Long-Term Statisticscited by 20
  3. 2025TelecomTS: A Multi-Modal Observability Dataset for Time Series and Language Analysiscited by 5
  4. 2025One-Embedding-Fits-All: Efficient Zero-Shot Time Series Forecasting by a Model Zoocited by 3
  5. 2025Why Cannot Neural Networks Master Extrapolation? Insights from Physical Lawscited by 2
  6. 2026Adapting Time Series Foundation Models through Data Mixturescited by 2
  7. 2025BinConv: A Neural Architecture for Ordinal Encoding in Time-Series Forecastingcited by 0
  8. 2026CloudCons: A Comprehensive End-to-End Benchmark for Cloud Resource Consolidationcited by 0
  9. 2026A Later Test Set Is Not a New Domain: Pretraining Familiarity Survives a Contamination-Free Hold-Outcited by 0
  10. 2026Evaluating Accuracy and Probabilistic Reliability of Zero-Shot Time Series Foundation Modelscited by 0