mlsanityData to 14 Oct 2026

Pre-trained Forecasting Models: Strong Zero-Shot Feature Extractors for Time Series Classification

arXiv 2510.26777 ↗First posted 30 Oct 2025cs.LGcited by 9

Not on the map. Why?

Abstract (arXiv)

Recent research on time series foundation models has primarily focused on forecasting, leaving it unclear how generalizable their learned representations are. In this study, we examine whether frozen pre-trained forecasting models can provide effective representations for classification. To this end, we compare different representation extraction strategies and introduce two model-agnostic embedding augmentations. Our experiments show that the best forecasting models achieve classification accuracy that matches or even surpasses that of state-of-the-art models pre-trained specifically for classification. Moreover, we observe a positive correlation between forecasting and classification performance. These findings challenge the assumption that task-specific pre-training is necessary, and suggest that learning to forecast may provide a powerful route toward constructing general-purpose time series foundation models.

Key facts. Cited by 9 papers in the map (data through 14 Oct 2026). Builds on 32 of its 36 references that are in the map. First posted on arXiv 30 Oct 2025.

Builds on

32 of 36 references.

References · 32

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2021On the Opportunities and Risks of Foundation Modelscited by 4,033
  3. 2022A Time Series is Worth 64 Words: Long-term Forecasting with Transformerscited by 2,039
  4. 2014How transferable are features in deep neural networks?cited by 1,225
  5. 2024Chronos: Learning the Language of Time Seriescited by 759
  6. 2023A decoder-only foundation model for time-series forecastingcited by 692
  7. 2023One Fits All:Power General Time Series Analysis by Pretrained LMcited by 632
  8. 2024Unified Training of Universal Time Series Forecasting Transformerscited by 614
  9. 2023Large Language Models Are Zero-Shot Time Series Forecasterscited by 500
  10. 2024MOMENT: A Family of Open Time-series Foundation Modelscited by 482
  11. 2021TS2Vec: Towards Universal Representation of Time Seriescited by 471
  12. 2018The UEA multivariate time series classification archive, 2018cited by 338
20 more
  1. 2024xLSTM: Extended Long Short-Term Memorycited by 327
  2. 2019Unsupervised Scalable Representation Learning for Multivariate Time Seriescited by 269
  3. 2022Self-Supervised Contrastive Pre-Training For Time Series via Time-Frequency Consistencycited by 250
  4. 2024GIFT-Eval: A Benchmark For General Time Series Forecasting Model Evaluationcited by 212
  5. 2025Sundial: A Family of Highly Capable Time Series Foundation Modelscited by 180
  6. 2020Benchmarking Multivariate Time Series Classification Algorithmscited by 145
  7. 2024Tiny Time Mixers (TTMs): Fast Pre-trained Models for Enhanced Zero/Few-Shot Forecasting of Multivariate Time Seriescited by 127
  8. 2025TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learningcited by 120
  9. 2023Lag-Llama: Towards Foundation Models for Probabilistic Time Series Forecastingcited by 113
  10. 2023Bake off redux: a review and experimental evaluation of recent time series classification algorithmscited by 101
  11. 2024UniTS: A Unified Multi-Task Time Series Modelcited by 94
  12. 2025This Time is Different: An Observability Perspective on Time Series Foundation Modelscited by 68
  13. 2023Ti-MAE: Self-Supervised Masked Time Series Autoencoderscited by 63
  14. 2024aeon: a Python toolkit for learning from time seriescited by 47
  15. 2025Mantis: Lightweight Foundation Model for Time Series Classificationcited by 24
  16. 2025Output Scaling: YingLong-Delayed Chain of Thought in a Large Pretrained Time Series Forecasting Modelcited by 24
  17. 2023NuTime: Numerically Multi-Scaled Embedding for Large-Scale Time-Series Pretrainingcited by 20
  18. 2024MIM-Refiner: A Contrastive Learning Boost from Intermediate Pre-Trained Representationscited by 19
  19. 2025xLSTM 7B: A Recurrent LLM for Fast and Efficient Inferencecited by 18
  20. 2025Zero-Shot Time Series Forecasting with Covariates via In-Context Learningcited by 12

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 1; 2026: 8.

9 papers.

  1. 2026Reverso: Efficient Time Series Foundation Models for Zero-shot Forecastingcited by 13
  2. 2026MantisV2: Closing the Zero-Shot Gap in Time Series Classification with Synthetic Data and Test-Time Strategiescited by 11
  3. 2026ARFBench: Benchmarking Time Series Question Answering Ability for Software Incident Responsecited by 6
  4. 2026Rethinking Zero-Shot Time Series Classification: From Task-specific Classifiers to In-Context Inferencecited by 3
  5. 2026Aionoscope: Debugging Latent-State Accessibility in Time-Series Representationscited by 2
  6. 2026LLM-based Agents for Forecasting and Prediction: Methods, Training, Evaluation, and Applicationscited by 1
  7. 2025TSFM in-context learning for time-series classification of bearing-health statuscited by 0
  8. 2026TimEE: End-to-end Time Series Classification via In-Context Learningcited by 0
  9. 2026When, Not How Much: Evaluating Time-Series Foundation Models on Sparse Eventscited by 0