pith. sign in

arxiv: 2605.17250 · v1 · pith:SZGADL3Vnew · submitted 2026-05-17 · 💻 cs.LG

Towards Principled Test-Time Adaptation for Time Series Forecasting

Haochun Wang , Ruichen Xu , Georgios Kementzidis , Karen Cho , Sebastian Ramirez Villarreal , Yuefan Deng This is my paper

Pith reviewed 2026-05-20 14:16 UTC · model grok-4.3

classification 💻 cs.LG
keywords test-time adaptationtime series forecastingfrequency domain analysisdistribution shiftlightweight calibrationprediction correctionforecasting under shift
0
0 comments X

The pith

A frequency-aware calibration method improves test-time adaptation for time series forecasting with far fewer parameters.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes a cleaner protocol for test-time adaptation in time series forecasting that relies only on matured ground truth values. It then shows through frequency analysis that prior adapters make only small and unstructured adjustments to the predictions. Building on this, it introduces Frequency-Aware Calibration, which directly adjusts the frequency components of the forecast. A reader would care because this yields an efficient, parameter-light way to handle distribution shifts in forecasting without retraining the base model.

Core claim

Under a protocol using only matured ground truth for adaptation, frequency-domain diagnosis indicates that existing TSF-TTA methods yield limited and weakly structured spectral modifications. Frequency-Aware Calibration addresses this by parameterizing corrections directly in the frequency domain, delivering competitive performance across datasets, horizons, and source forecasters while using substantially fewer trainable parameters.

What carries the argument

Frequency-Aware Calibration (FAC), which directly parameterizes prediction corrections in the frequency domain to produce more structured spectral adjustments.

If this is right

  • Test-time adaptation protocols can be simplified to use only matured ground truth.
  • Frequency domain offers a more effective space for calibrating forecasts than time domain methods.
  • Lightweight adapters with fewer parameters can match heavier ones in performance under distribution shift.
  • Consistent results hold across diverse datasets and forecasting horizons.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the frequency diagnosis holds, it could guide design of adapters for other time series tasks such as anomaly detection.
  • Reducing parameter count opens the door to on-device adaptation for streaming forecasting applications.
  • Similar frequency-based calibration might extend to handling shifts in multivariate time series or long-horizon predictions.

Load-bearing premise

Existing adapters only produce limited and weakly structured spectral modifications in their prediction corrections.

What would settle it

A frequency analysis of prediction corrections from existing adapters on additional datasets or forecasters that reveals strong structured spectral changes would undermine the need for a frequency-aware approach.

Figures

Figures reproduced from arXiv: 2605.17250 by Georgios Kementzidis, Haochun Wang, Karen Cho, Ruichen Xu, Sebastian Ramirez Villarreal, Yuefan Deng.

Figure 1
Figure 1. Figure 1: Comparison of adaptation supervision protocols in TSF-TTA. (a) Mixed supervision uses POGT [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison between the adjusted prediction of the first sample and later direct predictions within [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of Frequency-Aware Calibration (FAC) under the proposed protocol. FAC uses input and [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Frequency-domain magnitude of prediction corrections under the matured-only protocol for fore [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
read the original abstract

Test-time adaptation (TTA) has recently emerged as a promising approach for improving time series forecasting (TSF) under distribution shift. Existing TSF-TTA methods differ in how they utilize revealed targets, yet the resulting adaptation protocols remain heterogeneous and lack a clearly unified formulation. To address this issue, we revisit TSF-TTA from the perspective of protocol cleanliness and propose an adaptation protocol based solely on matured ground truth, yielding a more principled setting for adaptation. Under this protocol, we further diagnose existing adapters in the frequency domain and find that their prediction corrections often exhibit limited and weakly structured spectral modifications. Motivated by this diagnosis, we propose Frequency-Aware Calibration (FAC), a lightweight calibration method that directly parameterizes prediction corrections in the frequency domain. Across diverse datasets, forecasting horizons, and source forecasters, FAC achieves competitive and consistent performance while requiring substantially fewer trainable parameters than the compared TSF-TTA adapters.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a clean test-time adaptation (TTA) protocol for time series forecasting (TSF) that relies solely on matured ground-truth targets. It diagnoses existing TSF-TTA adapters via frequency-domain analysis of their prediction corrections, finding limited and weakly structured spectral modifications. Motivated by this, the authors introduce Frequency-Aware Calibration (FAC), a lightweight adapter that directly parameterizes corrections in the frequency domain. Across multiple datasets, horizons, and source forecasters, FAC is reported to match or exceed the performance of prior TTA methods while using substantially fewer trainable parameters.

Significance. If the frequency-domain diagnosis generalizes and the empirical comparisons are robust, FAC offers a more parameter-efficient and conceptually cleaner route to TTA in TSF. The emphasis on protocol cleanliness and the explicit frequency-domain parameterization could influence future adapter design in distribution-shift settings for sequential data.

major comments (2)
  1. [§3] §3 (Frequency-Domain Diagnosis): The central motivation for FAC rests on the claim that existing adapters produce only limited and weakly structured spectral modifications. This observation is used to justify direct frequency-domain parameterization, yet the analysis appears confined to the specific datasets, forecasters, and shift types examined; additional experiments on out-of-distribution shift families or alternative architectures would be needed to establish that the pattern is not an artifact of the chosen evaluation suite.
  2. [§5] §5 (Experiments): The abstract and results claim competitive performance with markedly fewer parameters, but the reported tables lack error bars, statistical significance tests, and full ablation studies on the frequency parameterization choices (e.g., number of frequency bins, regularization). Without these, it is difficult to determine whether the reported gains are stable or sensitive to post-hoc implementation decisions in the baselines.
minor comments (2)
  1. [§4] Notation for the frequency-domain correction operator should be introduced once and used consistently; currently the same symbol appears to be overloaded between the diagnosis and the FAC formulation.
  2. [Figure 2] Figure captions for the spectral plots should explicitly state the number of runs averaged and the exact frequency binning used.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and describe the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§3] §3 (Frequency-Domain Diagnosis): The central motivation for FAC rests on the claim that existing adapters produce only limited and weakly structured spectral modifications. This observation is used to justify direct frequency-domain parameterization, yet the analysis appears confined to the specific datasets, forecasters, and shift types examined; additional experiments on out-of-distribution shift families or alternative architectures would be needed to establish that the pattern is not an artifact of the chosen evaluation suite.

    Authors: We agree that broader validation would strengthen the claim. Our diagnosis was conducted on a diverse set of real-world datasets (ETT, Electricity, Traffic, Weather) and multiple source forecasters (Informer, Autoformer, FEDformer, DLinear) under standard distribution shifts in TSF. The limited spectral modifications were consistent across these settings. To address the concern, we will add experiments using synthetic distribution shifts (e.g., frequency-specific perturbations) and at least one additional architecture in the revised version. revision: yes

  2. Referee: [§5] §5 (Experiments): The abstract and results claim competitive performance with markedly fewer parameters, but the reported tables lack error bars, statistical significance tests, and full ablation studies on the frequency parameterization choices (e.g., number of frequency bins, regularization). Without these, it is difficult to determine whether the reported gains are stable or sensitive to post-hoc implementation decisions in the baselines.

    Authors: We acknowledge the value of these additions for demonstrating robustness. In the revision we will report mean and standard deviation over multiple random seeds, include statistical significance tests (e.g., paired t-tests against baselines), and expand the ablation study to vary the number of frequency bins and regularization strength, showing that performance remains stable within reasonable ranges. revision: yes

Circularity Check

0 steps flagged

Derivation chain is self-contained with empirical motivation and results

full rationale

The paper revisits TSF-TTA under a protocol based on matured ground truth, performs a frequency-domain diagnosis of existing adapters showing limited/weakly structured modifications, and proposes FAC as a direct parameterization in the frequency domain. All performance claims are framed as empirical outcomes from experiments across datasets, horizons, and source forecasters rather than quantities defined by or equivalent to the method's own fitted parameters. No load-bearing steps reduce to self-definition, fitted inputs renamed as predictions, or self-citation chains; the motivation is presented as an observation from analysis of prior methods, not a tautology. The derivation remains independent of its own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claims rest on the assumption that frequency-domain analysis of prediction corrections is a reliable diagnostic and that the new protocol does not introduce hidden biases. No explicit free parameters or invented entities are named in the abstract.

axioms (1)
  • domain assumption Existing TSF-TTA adapters produce limited and weakly structured spectral modifications under the matured-ground-truth protocol.
    This diagnosis is used to motivate the design of FAC.

pith-pipeline@v0.9.0 · 5699 in / 1248 out tokens · 32978 ms · 2026-05-20T14:16:18.935141+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

57 extracted references · 57 canonical work pages

  1. [3]

    So TTA : Robust test-time adaptation on noisy data streams

    Taesik Gong, Yewon Kim, Taeckyung Lee, Sorn Chottananurak, and Sung-Ju Lee. So TTA : Robust test-time adaptation on noisy data streams. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=3bdXag2rUd

    work page 2023

  2. [5]

    Shift-aware test time adaptation and benchmarking for time-series forecasting

    Shivam Grover and Ali Etemad. Shift-aware test time adaptation and benchmarking for time-series forecasting. In Second Workshop on Test-Time Adaptation: Putting Updates to the Test! at ICML 2025, 2025. URL https://openreview.net/forum?id=a399SmgWGl

    work page 2025

  3. [6]

    Online time series prediction using feature adjustment

    Xiannan Huang, Shuhan Qiu, Jiayuan Du, and Chao Yang. Online time series prediction using feature adjustment. In The Fourteenth International Conference on Learning Representations, 2026. URL https://openreview.net/forum?id=s4U2FWEMTU

    work page 2026

  4. [7]

    COSA : Context-aware output-space adapter for test-time adaptation in time series forecasting

    Jeonghwan Im and Hyuk-Yoon Kwon. COSA : Context-aware output-space adapter for test-time adaptation in time series forecasting. In The Fourteenth International Conference on Learning Representations, 2026. URL https://openreview.net/forum?id=L7Z5wBMPrW

    work page 2026

  5. [8]

    Buffer layers for test-time adaptation

    Hyeongyu Kim, GeonHui Han, and Dosik Hwang. Buffer layers for test-time adaptation. In The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2026. URL https://openreview.net/forum?id=sSZ9OM08KT

    work page 2026

  6. [10]

    Reversible instance normalization for accurate time-series forecasting against distribution shift

    Taesung Kim, Jinhee Kim, Yunwon Tae, Cheonbok Park, Jang-Ho Choi, and Jaegul Choo. Reversible instance normalization for accurate time-series forecasting against distribution shift. In International Conference on Learning Representations, 2022. URL https://openreview.net/forum?id=cGDAkQo1C0p

    work page 2022

  7. [11]

    Koa, Jan Chen, Yunshan Ma, Huanhuan Zheng, and Tat-Seng Chua

    Kelvin J.L. Koa, Jan Chen, Yunshan Ma, Huanhuan Zheng, and Tat-Seng Chua. Reasoning on time-series for financial technical analysis. In The Fourteenth International Conference on Learning Representations, 2026. URL https://openreview.net/forum?id=PcjIe5xNaY

    work page 2026

  8. [12]

    Fast and slow streams for online time series forecasting without information leakage

    Ying-yee Ava Lau, Zhiwen Shao, and Dit-Yan Yeung. Fast and slow streams for online time series forecasting without information leakage. In The Thirteenth International Conference on Learning Representations, 2025. URL https://openreview.net/forum?id=I0n3EyogMi

    work page 2025

  9. [14]

    A comprehensive survey on test-time adaptation under distribution shifts

    Jian Liang, Ran He, and Tieniu Tan. A comprehensive survey on test-time adaptation under distribution shifts. International Journal of Computer Vision, 133 0 (1): 0 31--64, 2025. doi:10.1007/s11263-024-02181-w. URL https://doi.org/10.1007/s11263-024-02181-w

    work page doi:10.1007/s11263-024-02181-w 2025

  10. [15]

    itransformer: Inverted transformers are effective for time series forecasting

    Yong Liu, Tengge Hu, Haoran Zhang, Haixu Wu, Shiyu Wang, Lintao Ma, and Mingsheng Long. itransformer: Inverted transformers are effective for time series forecasting. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=JePfAI8fah

    work page 2024

  11. [16]

    Adaptive normalization for non-stationary time series forecasting: A temporal slice perspective

    Zhiding Liu, Mingyue Cheng, Zhi Li, Zhenya Huang, Qi Liu, Yanhu Xie, and Enhong Chen. Adaptive normalization for non-stationary time series forecasting: A temporal slice perspective. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=5BqDSw8r5j

    work page 2023

  12. [17]

    Future energy insights: Time-series and deep learning models for city load forecasting

    Neda Maleki, Oxana Lundström, Arslan Musaddiq, John Jeansson, Tobias Olsson, and Fredrik Ahlgren. Future energy insights: Time-series and deep learning models for city load forecasting. Applied Energy, 374: 0 124067, 2024. doi:10.1016/j.apenergy.2024.124067. URL https://www.sciencedirect.com/science/article/pii/S0306261924014508

    work page doi:10.1016/j.apenergy.2024.124067 2024

  13. [18]

    Oliveira, and Saghar Irandoust

    Heitor Rapela Medeiros, Hossein Sharifi-Noghabi, Gabriel L. Oliveira, and Saghar Irandoust. Accurate parameter-efficient test-time adaptation for time series forecasting. In Second Workshop on Test-Time Adaptation: Putting Updates to the Test! at ICML 2025, 2025. URL https://openreview.net/forum?id=uFj4EL4GTB

    work page 2025

  14. [19]

    A time series is worth 64 words: Long-term forecasting with transformers

    Yuqi Nie, Nam H Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam. A time series is worth 64 words: Long-term forecasting with transformers. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=Jbdc0vTOcol

    work page 2023

  15. [20]

    Learning fast and slow for online time series forecasting

    Quang Pham, Chenghao Liu, Doyen Sahoo, and Steven Hoi. Learning fast and slow for online time series forecasting. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=q-PbpHD3EOk

    work page 2023

  16. [21]

    An analysis of linear time series forecasting models

    William Toner and Luke Nicholas Darlow. An analysis of linear time series forecasting models. In Forty-first International Conference on Machine Learning, 2024. URL https://openreview.net/forum?id=xl82CcbYaT

    work page 2024

  17. [22]

    Ruey S. Tsay. Analysis of Financial Time Series. John Wiley & Sons, 3 edition, 2010. ISBN 9780470414354. doi:10.1002/9780470644560

    work page doi:10.1002/9780470644560 2010

  18. [23]

    Clim ODE : Climate and weather forecasting with physics-informed neural ODE s

    Yogesh Verma, Markus Heinonen, and Vikas Garg. Clim ODE : Climate and weather forecasting with physics-informed neural ODE s. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=xuY33XhEGR

    work page 2024

  19. [24]

    Tent: Fully test-time adaptation by entropy minimization

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=uXl3bZLkr3c

    work page 2021

  20. [25]

    MICN : Multi-scale local and global context modeling for long-term series forecasting

    Huiqiang Wang, Jian Peng, Feihu Huang, Jince Wang, Junhui Chen, and Yifei Xiao. MICN : Multi-scale local and global context modeling for long-term series forecasting. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=zt53IDUR1U

    work page 2023

  21. [26]

    Masked feature prediction for self-supervised visual pre-training

    Qin Wang, Olga Fink, Luc Van Gool, and Dengxin Dai. Continual test-time domain adaptation. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 7191--7201, 2022. doi:10.1109/CVPR52688.2022.00706

    work page doi:10.1109/cvpr52688.2022.00706 2022

  22. [27]

    Interpretable weather forecasting for worldwide stations with a unified deep model

    Haixu Wu, Hang Zhou, Mingsheng Long, and Jianmin Wang. Interpretable weather forecasting for worldwide stations with a unified deep model. Nature Machine Intelligence, 5 0 (6): 0 602--611, 2023. doi:10.1038/s42256-023-00667-9. URL https://doi.org/10.1038/s42256-023-00667-9

    work page doi:10.1038/s42256-023-00667-9 2023

  23. [28]

    Frequency-domain MLP s are more effective learners in time series forecasting

    Kun Yi, Qi Zhang, Wei Fan, Shoujin Wang, Pengyang Wang, Hui He, Ning An, Defu Lian, Longbing Cao, and Zhendong Niu. Frequency-domain MLP s are more effective learners in time series forecasting. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=iif9mGCTfy

    work page 2023

  24. [29]

    Deep learning on traffic prediction: Methods, analysis, and future directions

    Xueyan Yin, Genze Wu, Jinze Wei, Yanming Shen, Heng Qi, and Baocai Yin. Deep learning on traffic prediction: Methods, analysis, and future directions. IEEE Transactions on Intelligent Transportation Systems, 23 0 (6), 2022

    work page 2022

  25. [32]

    Onenet: Enhancing time series forecasting models under concept drift by online ensembling

    Yi-Fan Zhang, Qingsong Wen, Xue Wang, Weiqi Chen, Liang Sun, Zhang Zhang, Liang Wang, Rong Jin, and Tieniu Tan. Onenet: Enhancing time series forecasting models under concept drift by online ensembling. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=Q25wMXsaeZ

    work page 2023

  26. [33]

    2026 , url=

    Jeonghwan Im and Hyuk-Yoon Kwon , booktitle=. 2026 , url=

    work page 2026

  27. [34]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Battling the Non-stationarity in Time Series Forecasting via Test-time Adaptation , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2025 , month=. doi:10.1609/aaai.v39i17.33965 , number=

    work page doi:10.1609/aaai.v39i17.33965 2025

  28. [35]

    Interpretable weather forecasting for worldwide stations with a unified deep model , volume =

    Wu, Haixu and Zhou, Hang and Long, Mingsheng and Wang, Jianmin , journal =. Interpretable weather forecasting for worldwide stations with a unified deep model , volume =. 2023 , doi =

    work page 2023

  29. [36]

    Yogesh Verma and Markus Heinonen and Vikas Garg , booktitle=. Clim. 2024 , url=

    work page 2024

  30. [37]

    Deep Learning on Traffic Prediction: Methods, Analysis, and Future Directions , volume =

    Yin, Xueyan and Wu, Genze and Wei, Jinze and Shen, Yanming and Qi, Heng and Yin, Baocai , journal =. Deep Learning on Traffic Prediction: Methods, Analysis, and Future Directions , volume =

    work page

  31. [38]

    2024 , isbn =

    Zhang, Weijia and Zhang, Le and Han, Jindong and Liu, Hao and Fu, Yanjie and Zhou, Jingbo and Mei, Yu and Xiong, Hui , title =. 2024 , isbn =. doi:10.1145/3637528.3671665 , booktitle =

    work page doi:10.1145/3637528.3671665 2024

  32. [39]

    Future energy insights: Time-series and deep learning models for city load forecasting , volume =

    Maleki, Neda and Lundström, Oxana and Musaddiq, Arslan and Jeansson, John and Olsson, Tobias and Ahlgren, Fredrik , journal =. Future energy insights: Time-series and deep learning models for city load forecasting , volume =. 2024 , url =

    work page 2024

  33. [40]

    The Fourteenth International Conference on Learning Representations , year=

    Reasoning on Time-Series for Financial Technical Analysis , author=. The Fourteenth International Conference on Learning Representations , year=

    work page

  34. [41]

    International Conference on Learning Representations , year=

    Reversible Instance Normalization for Accurate Time-Series Forecasting against Distribution Shift , author=. International Conference on Learning Representations , year=

    work page

  35. [42]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Dish-TS: A General Paradigm for Alleviating Distribution Shift in Time Series Forecasting , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2023 , month=. doi:10.1609/aaai.v37i6.25914 , number=

    work page doi:10.1609/aaai.v37i6.25914 2023

  36. [43]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    Adaptive Normalization for Non-stationary Time Series Forecasting: A Temporal Slice Perspective , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

    work page

  37. [44]

    The Thirteenth International Conference on Learning Representations , year=

    Fast and Slow Streams for Online Time Series Forecasting Without Information Leakage , author=. The Thirteenth International Conference on Learning Representations , year=

    work page

  38. [45]

    arXiv preprint arXiv:2412.00108 , year =

    Act Now: A Novel Online Forecasting Framework for Large-Scale Streaming Data , author =. arXiv preprint arXiv:2412.00108 , year =

    work page arXiv

  39. [46]

    The Fourteenth International Conference on Learning Representations , year=

    Online time series prediction using feature adjustment , author=. The Fourteenth International Conference on Learning Representations , year=

    work page

  40. [47]

    The Eleventh International Conference on Learning Representations , year=

    Learning Fast and Slow for Online Time Series Forecasting , author=. The Eleventh International Conference on Learning Representations , year=

    work page

  41. [48]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    OneNet: Enhancing Time Series Forecasting Models under Concept Drift by Online Ensembling , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

    work page

  42. [49]

    Second Workshop on Test-Time Adaptation: Putting Updates to the Test! at ICML 2025 , year=

    Accurate Parameter-Efficient Test-Time Adaptation for Time Series Forecasting , author=. Second Workshop on Test-Time Adaptation: Putting Updates to the Test! at ICML 2025 , year=

    work page 2025

  43. [50]

    Second Workshop on Test-Time Adaptation: Putting Updates to the Test! at ICML 2025 , year=

    Shift-Aware Test Time Adaptation and Benchmarking for Time-Series Forecasting , author=. Second Workshop on Test-Time Adaptation: Putting Updates to the Test! at ICML 2025 , year=

    work page 2025

  44. [51]

    A Comprehensive Survey on Test-Time Adaptation Under Distribution Shifts , volume =

    Liang, Jian and He, Ran and Tan, Tieniu , journal =. A Comprehensive Survey on Test-Time Adaptation Under Distribution Shifts , volume =. 2025 , doi =

    work page 2025

  45. [52]

    International Conference on Learning Representations , year=

    Tent: Fully Test-Time Adaptation by Entropy Minimization , author=. International Conference on Learning Representations , year=

    work page

  46. [53]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    Buffer layers for Test-Time Adaptation , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    work page

  47. [54]

    Taesik Gong and Yewon Kim and Taeckyung Lee and Sorn Chottananurak and Sung-Ju Lee , booktitle=. So. 2023 , url=

    work page 2023

  48. [55]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Are Transformers Effective for Time Series Forecasting? , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2023 , month=. doi:10.1609/aaai.v37i9.26317 , number=

    work page doi:10.1609/aaai.v37i9.26317 2023

  49. [56]

    The Twelfth International Conference on Learning Representations , year=

    iTransformer: Inverted Transformers Are Effective for Time Series Forecasting , author=. The Twelfth International Conference on Learning Representations , year=

    work page

  50. [57]

    The Eleventh International Conference on Learning Representations , year=

    A Time Series is Worth 64 Words: Long-term Forecasting with Transformers , author=. The Eleventh International Conference on Learning Representations , year=

    work page

  51. [58]

    Frequency-domain

    Kun Yi and Qi Zhang and Wei Fan and Shoujin Wang and Pengyang Wang and Hui He and Ning An and Defu Lian and Longbing Cao and Zhendong Niu , booktitle=. Frequency-domain. 2023 , url=

    work page 2023

  52. [59]

    Forty-first International Conference on Machine Learning , year=

    An Analysis of Linear Time Series Forecasting Models , author=. Forty-first International Conference on Machine Learning , year=

    work page

  53. [60]

    2023 , url=

    Huiqiang Wang and Jian Peng and Feihu Huang and Jince Wang and Junhui Chen and Yifei Xiao , booktitle=. 2023 , url=

    work page 2023

  54. [61]

    2010 , doi =

    Analysis of Financial Time Series , author =. 2010 , doi =

    work page 2010

  55. [62]

    Continual Test-Time Domain Adaptation , year=

    Wang, Qin and Fink, Olga and Van Gool, Luc and Dai, Dengxin , booktitle=. Continual Test-Time Domain Adaptation , year=

    work page

  56. [63]

    2021 , isbn =

    Du, Yuntao and Wang, Jindong and Feng, Wenjie and Pan, Sinno and Qin, Tao and Xu, Renjun and Wang, Chongjun , title =. 2021 , isbn =. doi:10.1145/3459637.3482315 , booktitle =

    work page doi:10.1145/3459637.3482315 2021

  57. [64]

    Granger, Clive W. J. , title =. Oxford Bulletin of Economics and Statistics , volume =. doi:10.1046/j.0305-9049.2003.00094.x , url =

    work page doi:10.1046/j.0305-9049.2003.00094.x 2003