pith. sign in

arxiv: 2606.07725 · v1 · pith:M47DCPK2new · submitted 2026-06-05 · ⚛️ physics.geo-ph · cs.LG

GNSS-FM: A Self-Supervised Foundation Model for Daily GNSS Displacement Time Series

Nick Teutschmann (1) , Laura Crocetti (1) , Fanny Lehmann (2) , Leonardo Trentini (1) , Benedikt Soja (1) ((1) Institute of Geodesy , Photogrammetry , ETH Zurich , Switzerland
show 2 more authors
(2) ETH AI Center Switzerland)
This is my paper

Pith reviewed 2026-06-27 20:06 UTC · model grok-4.3

classification ⚛️ physics.geo-ph cs.LG
keywords GNSSself-supervised learningfoundation modeldisplacement time seriesseismic detectiontectonic monitoringtime series forecasting
0
0 comments X

The pith

Self-supervised pretraining on unlabeled GNSS displacement data produces representations that improve performance on forecasting and seismic localization tasks.

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

The paper presents GNSS-FM, a foundation model for daily GNSS time series that is pretrained in a self-supervised manner on data from over 17,000 stations. It combines displacement and velocity-like increments as input and uses a masked latent prediction objective with vector-quantized targets. The model is then fine-tuned for 90-day displacement forecasting and seismic step localization, outperforming task-specific baselines in both. This approach addresses the scarcity of labeled data by leveraging abundant unlabeled GNSS observations for applications in tectonic monitoring and earthquake cycle studies.

Core claim

The authors claim that pretraining with self-supervised masked prediction on global GNSS data allows the model to learn representations that capture seismic offsets, tectonic drift, and seasonal patterns, leading to better results when adapted to specific tasks like forecasting displacements over 90 days and detecting seismic steps compared to models trained from scratch on those tasks.

What carries the argument

Dual-stream input of displacement and velocity increments with a masked latent prediction objective using vector-quantized targets adapted from wav2vec 2.0 for geodetic time series.

If this is right

  • The foundation model captures main signal types including seismic offsets, tectonic drift, and seasonal patterns.
  • Fine-tuning leads to better 90-day displacement forecasting than strong task-specific baselines.
  • Fine-tuning leads to better seismic step localization than strong task-specific baselines.
  • Self-supervised pretraining is a promising approach for GNSS time series analysis.

Where Pith is reading between the lines

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

  • Similar self-supervised models could be developed for other types of geophysical time series data.
  • The learned codebook could enable unsupervised detection of geophysical events beyond the fine-tuned tasks.
  • Pretraining on global data might allow better generalization to regions with sparse labeled data.

Load-bearing premise

The representations learned during pretraining capture the main signal types in GNSS displacement data, including seismic offsets, tectonic drift, and seasonal patterns.

What would settle it

Demonstrating that the fine-tuned GNSS-FM does not outperform task-specific baselines on either the forecasting or the seismic localization task would falsify the central claim.

Figures

Figures reproduced from arXiv: 2606.07725 by (2) ETH AI Center, Benedikt Soja (1) ((1) Institute of Geodesy, ETH Zurich, Fanny Lehmann (2), Laura Crocetti (1), Leonardo Trentini (1), Nick Teutschmann (1), Photogrammetry, Switzerland, Switzerland).

Figure 1
Figure 1. Figure 1: Dataset overview. Left: Global distribution of GNSS stations in the geographically stratified training, validation, and test splits. Right: Example of a [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of GNSS-FM. The model takes displacement and velocity time series as inputs, together with station metadata. During self-supervised [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Quantizer usage over training. Left: Per-group hard-assignment [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Example occurrence of the event-sensitive code g3-c221 at station J300. The shaded region shows the receptive field of the latent position assigned to [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of the dual-stream model with the single-stream ablations. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Histogram of localization errors on windows that both contain a [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: Representative probabilistic forecast from GNSS-FM Stage 2 on the [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Representative seismic step localization for one example in the test [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Forecast example for station J263 illustrating the behavior of GNSS [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Localization performance on true-event windows in the test set, [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
read the original abstract

Displacement time series from Global Navigation Satellite Systems (GNSS) are essential for a wide range of applications, including monitoring tectonic crustal deformations and investigating the different stages of the earthquake cycle. Machine learning methods have proven promising for GNSS applications; however, most remain fully supervised. This creates a bottleneck as labeled data are scarce, even though large amounts of unlabeled GNSS data are freely available. We present GNSS-FM, a self-supervised foundation model for daily GNSS time series. The model uses a dual-stream input combining displacement and velocity-like increments, and is pretrained using a masked latent prediction objective with vector-quantized targets adapted from wav2vec 2.0, with several modifications for geodetic data. Pretrained on data from over 17,000 globally distributed GNSS stations, an analysis of the learned codebook suggests that the representations capture the main signal types in GNSS displacement data, including seismic offsets, tectonic drift, and seasonal patterns. The foundation model is later fine-tuned on two downstream tasks, namely 90-day displacement forecasting and seismic step localization, where it outperforms strong task-specific baselines in both cases. These results show that self-supervised pretraining is a promising approach for GNSS time series analysis.

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 / 1 minor

Summary. The manuscript introduces GNSS-FM, a self-supervised foundation model for daily GNSS displacement time series. It is pretrained on unlabeled data from over 17,000 globally distributed stations using a dual-stream (displacement and velocity-like increments) input and a masked latent prediction objective with vector-quantized targets adapted from wav2vec 2.0. An analysis of the learned codebook is presented as evidence that the representations capture key geophysical signals including seismic offsets, tectonic drift, and seasonal patterns. The pretrained model is then fine-tuned on two downstream tasks—90-day displacement forecasting and seismic step localization—where it is claimed to outperform strong task-specific baselines.

Significance. If the reported outperformance is shown to arise specifically from the self-supervised pretraining (rather than architecture or data scale alone), the work would demonstrate a viable path for leveraging abundant unlabeled GNSS data to improve performance on tasks where labeled examples remain scarce. The adaptation of wav2vec-style objectives to geodetic time series and the dual-stream design represent concrete technical contributions that could be adopted more broadly in the field.

major comments (2)
  1. [codebook analysis] Codebook analysis (abstract and associated results section): The assertion that the learned representations capture seismic offsets, tectonic drift, and seasonal patterns rests on an analysis described only as 'suggestive.' No quantitative mapping is supplied (e.g., precision/recall of codebook entries against an independent earthquake catalog for offsets, or reconstruction error conditioned on known tectonic/seasonal regimes). Without such metrics, the causal connection between pretraining and the claimed downstream gains remains unestablished; the observed improvements could equally result from model capacity or fine-tuning procedure.
  2. [results on downstream tasks] Downstream task evaluation (results section on 90-day forecasting and seismic step localization): The abstract states that fine-tuned GNSS-FM 'outperforms strong task-specific baselines in both cases,' yet supplies no numerical metrics, baseline specifications, statistical tests, confidence intervals, or ablation controls. These details are load-bearing for the central claim that self-supervised pretraining confers an advantage; their absence prevents assessment of whether the gains are robust or reproducible.
minor comments (1)
  1. [abstract] The abstract and introduction would benefit from explicit citation of the exact number of stations, total time span, and sampling rate used in pretraining to allow readers to gauge data scale.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below and will revise the manuscript to strengthen the evidence supporting our claims regarding the codebook analysis and downstream task evaluations.

read point-by-point responses

  1. Referee: [codebook analysis] Codebook analysis (abstract and associated results section): The assertion that the learned representations capture seismic offsets, tectonic drift, and seasonal patterns rests on an analysis described only as 'suggestive.' No quantitative mapping is supplied (e.g., precision/recall of codebook entries against an independent earthquake catalog for offsets, or reconstruction error conditioned on known tectonic/seasonal regimes). Without such metrics, the causal connection between pretraining and the claimed downstream gains remains unestablished; the observed improvements could equally result from model capacity or fine-tuning procedure.

    Authors: We agree that describing the codebook analysis as merely 'suggestive' does not sufficiently establish the link between the learned representations and specific geophysical signals, nor does it isolate the contribution of pretraining to downstream gains. In the revised manuscript, we will add quantitative evaluations: precision and recall metrics for codebook entries associated with seismic offsets by cross-referencing against an independent earthquake catalog; and regime-conditioned reconstruction or classification errors for tectonic drift and seasonal patterns. These additions will provide a more rigorous mapping and help demonstrate that the representations capture the relevant signals. revision: yes

  2. Referee: [results on downstream tasks] Downstream task evaluation (results section on 90-day forecasting and seismic step localization): The abstract states that fine-tuned GNSS-FM 'outperforms strong task-specific baselines in both cases,' yet supplies no numerical metrics, baseline specifications, statistical tests, confidence intervals, or ablation controls. These details are load-bearing for the central claim that self-supervised pretraining confers an advantage; their absence prevents assessment of whether the gains are robust or reproducible.

    Authors: We concur that the absence of specific numerical metrics, baseline details, statistical tests, confidence intervals, and ablation controls in the abstract and results section limits the ability to evaluate the robustness of the claims and the specific benefit of self-supervised pretraining. The revised manuscript will include: exact performance numbers (e.g., RMSE or MAE for 90-day forecasting and precision/recall or F1 for step localization); full specifications of the task-specific baselines including architectures and training protocols; results of statistical significance tests with confidence intervals; and ablation studies comparing the pretrained GNSS-FM against non-pretrained or randomly initialized counterparts to isolate the effect of the self-supervised pretraining from model capacity or fine-tuning alone. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical pretraining and fine-tuning chain is self-contained

full rationale

The paper describes standard self-supervised pretraining on unlabeled GNSS data (>17k stations) using a masked latent prediction objective adapted from wav2vec 2.0, followed by separate fine-tuning on two downstream tasks (90-day forecasting and seismic step localization) with reported outperformance versus task-specific baselines. No equations, definitions, or self-citations reduce any claimed result to its own inputs by construction. The codebook analysis is presented as a post-hoc observation rather than a load-bearing derivation. All performance claims rest on external empirical benchmarks, satisfying the self-contained criterion for a score of 0.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no information on free parameters, axioms, or invented entities used in the model.

pith-pipeline@v0.9.1-grok · 5789 in / 1228 out tokens · 25702 ms · 2026-06-27T20:06:12.383698+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

55 extracted references · 9 canonical work pages · 7 internal anchors

  1. [1]

    Geodynamics,

    J. Freymueller, “Geodynamics,” inSpringer Handbook of Global Nav- igation Satellite Systems, P. J. Teunissen and O. Montenbruck, Eds. Cham: Springer International Publishing, 2017

    2017

  2. [2]

    Harnessing the GPS data explosion for interdisciplinary science,

    G. Blewitt, W. C. Hammond, and C. Kreemer, “Harnessing the GPS data explosion for interdisciplinary science,”Eos, vol. 99, 2018

    2018

  3. [3]

    Discontinuity detection in GNSS station coordinate time series using machine learning,

    L. Crocetti, M. Schartner, and B. Soja, “Discontinuity detection in GNSS station coordinate time series using machine learning,”Remote Sensing, vol. 13, p. 3906, 2021

    2021

  4. [4]

    An improved VMD-LSTM model for time-varying GNSS time series prediction with temporally correlated noise,

    H. Chen, T. Lu, J. Huang, X. He, K. Yu, X. Sun, X. Ma, and Z. Huang, “An improved VMD-LSTM model for time-varying GNSS time series prediction with temporally correlated noise,”Remote Sensing, vol. 15, no. 14, p. 3694, 2023

    2023

  5. [5]

    Multi-station deep learning on geodetic time series detects slow slip events in cascadia,

    G. Costantino, S. Giffard-Roisin, M. Radiguet, M. Dalla Mura, D. Marsan, and A. Socquet, “Multi-station deep learning on geodetic time series detects slow slip events in cascadia,”Communications Earth & Environment, vol. 4, p. 435, 2023

    2023

  6. [6]

    BERT: Pre- training of deep bidirectional transformers for language understanding,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre- training of deep bidirectional transformers for language understanding,” inProceedings of NAACL-HLT, 2019

    2019

  7. [7]

    Masked autoencoders are scalable vision learners,

    K. He, X. Chen, S. Xie, Y . Li, P. Doll ´ar, and R. Girshick, “Masked autoencoders are scalable vision learners,” inProceedings of CVPR,

  8. [8]

    Masked Autoencoders Are Scalable Vision Learners

    [Online]. Available: https://arxiv.org/abs/2111.06377

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [9]

    wav2vec 2.0: A framework for self-supervised learning of speech representations,

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” inAdvances in Neural Information Processing Systems, vol. 33. Curran Associates, Inc., 2020, pp. 12 449– 12 460. [Online]. Available: https://proceedings.neurips.cc/paper/2020/ hash/92d1e1eb1cd6f9fba3227870bb6d7f07-Abstract.html

    2020

  10. [10]

    SeisLM: a foundation model for seismic waveforms,

    T. Liu, J. M ¨unchmeyer, L. Laurenti, C. Marone, M. V . d. Hoop, and I. Dokmani ´c, “SeisLM: a foundation model for seismic waveforms,” arXiv preprint arXiv:2410.15765, 2024

    work page arXiv 2024

  11. [11]

    Beit: Bert pre-training of image transformers,

    H. Bao, L. Dong, S. Piao, and F. Wei, “Beit: Bert pre-training of image transformers,” inInternational Conference on Learning Representations (ICLR), 2022

    2022

  12. [12]

    Feature guided masked autoencoder for self-supervised learning in remote sens- ing,

    Y . Wang, H. Hern ´andez Hern ´andez, C. Albrecht, and X. Zhu, “Feature guided masked autoencoder for self-supervised learning in remote sens- ing,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, pp. 1–17, 2024

    2024

  13. [13]

    HuBERT: Self-supervised speech representation learning by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdinov, and A. Mohamed, “HuBERT: Self-supervised speech representation learning by masked prediction of hidden units,”IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 29, pp. 3451–3460, 2021

    2021

  14. [14]

    WavLM: Large-scale self-supervised pre-training for full stack speech processing,

    S. Chen, C. Wang, Z. Chen, Y . Wu, S. Liu, Z. Chen, J. Li, Y . Qian, and F. Wei, “WavLM: Large-scale self-supervised pre-training for full stack speech processing,”IEEE Journal of Selected Topics in Signal Processing, vol. 16, no. 6, pp. 1505–1518, 2022

    2022

  15. [15]

    TS2Vec: Towards universal representation of time series,

    Z. Yue, Y . Wang, J. Duan, T. Yang, C. Huang, Y . Tong, and B. Xu, “TS2Vec: Towards universal representation of time series,” inProceed- ings of the AAAI Conference on Artificial Intelligence, vol. 36, no. 8, 2022, pp. 8980–8987

    2022

  16. [16]

    A transformer-based framework for multivariate time series representation learning,

    G. Zerveas, S. Jayaraman, D. Patel, A. Bhamidipaty, and C. Eickhoff, “A transformer-based framework for multivariate time series representation learning,” inProceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2021, pp. 2114–2124

    2021

  17. [17]

    Vector quantization pretraining for EEG time series with random projection and phase alignment,

    H. Gui, X. Li, and X. Chen, “Vector quantization pretraining for EEG time series with random projection and phase alignment,” in Proceedings of the 41st International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, R. Salakhutdinov, Z. Kolter, K. Heller, A. Weller, N. Oliver, J. Scarlett, and F. Berkenkamp, Eds., vol. 235. P...

    2024

  18. [18]

    Effective self-supervised transformers for sparse time series data,

    A. Labach, A. Pokhrel, S. E. Yi, S. Zuberi, M. V olkovs, and R. G. Krishnan, “Effective self-supervised transformers for sparse time series data,”OpenReview, 2022. [Online]. Available: https://openreview.net/forum?id=HUCgU5EQluN

    2022

  19. [19]

    Self-supervised spatio-temporal representation learning of satellite image time series,

    I. Dumeur, S. Valero, and J. Inglada, “Self-supervised spatio-temporal representation learning of satellite image time series,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 17, pp. 4350–4367, 2024

    2024

  20. [20]

    Chronos: Learning the Language of Time Series

    A. F. Ansariet al., “Chronos: Learning the language of time series,”arXiv preprint arXiv:2403.07815, 2024. [Online]. Available: https://arxiv.org/abs/2403.07815

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  21. [21]

    A decoder-only foundation model for time-series forecasting

    A. Das, W. Kong, R. Sen, and Y . Zhou, “A decoder-only foundation model for time-series forecasting,”arXiv preprint arXiv:2310.10688, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  22. [22]

    TimeGPT-1,

    A. Garza, C. Challu, and M. Mergenthaler-Canseco, “TimeGPT-1,”arXiv preprint arXiv:2310.03589, 2023

    work page arXiv 2023

  23. [23]

    A foundation model for the earth system,

    C. Bodnar, W. P. Bruinsma, A. Lucic, M. Stanley, A. Allen, J. Brand- stetter, P. Garvan, M. Riechert, J. A. Weyn, H. Dong, J. K. Gupta, K. Thambiratnam, A. T. Archibald, C.-C. Wu, E. Heider, M. Welling, R. E. Turner, and P. Perdikaris, “A foundation model for the earth system,”Nature, vol. 641, pp. 1180–1187, 2025

    2025

  24. [24]

    Integrating GNSS- derived atmospheric delays into large weather foundation models,

    L. Trentini, F. Lehmann, L. Crocetti, and B. Soja, “Integrating GNSS- derived atmospheric delays into large weather foundation models,” in EGU General Assembly 2026. Vienna, Austria: Copernicus Meetings, 2026, eGU26-6706

    2026

  25. [25]

    Earth system foundation model (esfm): A unified framework for heterogeneous data integration and forecasting,

    F. Ozdemir, Y . Cheng, S. Mohebi, F. Lehmann, S. Adamov, Z. Zhang, L. Trentini, D. Grund, O. Fuhrer, T. Hoefler, S. Mishra, S. Schemm, B. Soja, and M. Salzmann, “Earth system foundation model (esfm): A unified framework for heterogeneous data integration and forecasting,” 2026

    2026

  26. [26]

    Alphaearth foundations: An embedding field model for accurate and efficient global mapping from sparse label data,

    C. F. Brown, M. R. Kazmierski, V . J. Pasquarella, W. J. Ruck- lidge, M. Samsikova, C. Zhang, E. Shelhamer, E. Lahera, O. Wiles, S. Ilyushchenko, N. Gorelick, L. L. Zhang, S. Alj, E. Schechter, S. Askay, O. Guinan, R. Moore, A. Boukouvalas, and P. Kohli, “Alphaearth foundations: An embedding field model for accurate and efficient global mapping from spars...

    2025

  27. [27]

    Denoising daily displacement GNSS time series using deep neural networks in a near real-time framing: a single-station method,

    G. Mastella, J. Bedford, F. Corbi, and F. Funiciello, “Denoising daily displacement GNSS time series using deep neural networks in a near real-time framing: a single-station method,”Geophysical Journal Inter- national, vol. 242, no. 3, p. ggaf207, 2025

    2025

  28. [28]

    Cascadia daily GNSS time series denoising: Graph neural network and stack filtering,

    L. Bachelot, A. M. Thomas, D. Melgar, J. Searcy, and Y .-S. Sun, “Cascadia daily GNSS time series denoising: Graph neural network and stack filtering,”Seismica, vol. 2, no. 4, 2023

    2023

  29. [29]

    Modeling of residual GNSS station motions through meteorological data in a machine learning approach,

    P. Ruttner, R. Hohensinn, S. D’Aronco, J. D. Wegner, and B. Soja, “Modeling of residual GNSS station motions through meteorological data in a machine learning approach,”Remote Sensing, vol. 14, no. 1, p. 17, 2022

    2022

  30. [30]

    Correction models for GNSS displacements in europe based on en- vironmental variables and XGBoost,

    L. Crocetti, M. Schartner, R. Schneider, K. Schindler, and B. Soja, “Correction models for GNSS displacements in europe based on en- vironmental variables and XGBoost,”Earth and Space Science Open Archive, 2025, preprint

    2025

  31. [31]

    Modelling of GNSS station position time series using deep learning approaches,

    M. S ¸ims ¸ek, M. Tas ¸kıran, and U. Do˘gan, “Modelling of GNSS station position time series using deep learning approaches,”Earth Science Informatics, vol. 18, no. 1, p. 96, 2024

    2024

  32. [32]

    Kiani Shahvandi and B

    M. Kiani Shahvandi and B. Soja, “Inclusion of data uncertainty in machine learning and its application in geodetic data science, with case studies for the prediction of earth orientation parameters and GNSS station coordinate time series,”Advances in Space Research, vol. 70, no. 3, pp. 563–575, 2022

    2022

  33. [33]

    The effect of coloured noise on the uncertainties of rates estimated from geodetic time series,

    S. D. P. Williams, “The effect of coloured noise on the uncertainties of rates estimated from geodetic time series,”Journal of Geodesy, vol. 76, no. 9, pp. 483–494, 2003

    2003

  34. [34]

    Anatomy of apparent seasonal variations from GPS-derived site position time series,

    D. Dong, P. Fang, Y . Bock, M. K. Cheng, and S. Miyazaki, “Anatomy of apparent seasonal variations from GPS-derived site position time series,” Journal of Geophysical Research, vol. 107, no. B4, 2002. 14

    2002

  35. [35]

    Itrf2020: an augmented reference frame refining the modeling of nonlinear station motions,

    Z. Altamimi, P. Rebischung, X. Collilieux, L. M ´etivier, and K. Chanard, “Itrf2020: an augmented reference frame refining the modeling of nonlinear station motions,”Journal of Geodesy, vol. 97, no. 47, 2023

    2023

  36. [36]

    Wdowinski, Y

    S. Wdowinski, Y . Bock, J. Zhang, P. Fang, and J. Genrich, “Southern California permanent GPS geodetic array: Spatial filtering of daily posi- tions for estimating coseismic and postseismic displacements induced by the 1992 Landers earthquake,”Journal of Geophysical Research: Solid Earth, vol. 102, no. B8, pp. 18 057–18 070, 1997

    1992

  37. [37]

    Spa- tiotemporal filtering using principal component analysis and karhunen– lo`eve expansion approaches to GPS coordinate time series,

    D. Dong, P. Fang, Y . Bock, M. K. Cheng, and S. Miyazaki, “Spa- tiotemporal filtering using principal component analysis and karhunen– lo`eve expansion approaches to GPS coordinate time series,”Journal of Geophysical Research: Solid Earth, vol. 111, no. B3, 2006

    2006

  38. [38]

    Noise-resilient GNSS coordinate time series prediction using A VMD-sLSTM-transformer hybrid model,

    J. Jiao, H. Wang, Y . Dang, Y . Ren, C. Yue, X. Wu, H. Cui, and X. Wang, “Noise-resilient GNSS coordinate time series prediction using A VMD-sLSTM-transformer hybrid model,”Advances in Space Research, vol. 76, no. 11, pp. 6863–6881, 2025

    2025

  39. [39]

    Generalized hampel filters,

    R. K. Pearson, Y . Neuvo, J. Astola, and M. Gabbouj, “Generalized hampel filters,”EURASIP Journal on Advances in Signal Processing, vol. 2016, p. 87, 2016

    2016

  40. [40]

    The identification of multiple outliers,

    L. Davies and U. Gather, “The identification of multiple outliers,” Journal of the American Statistical Association, vol. 88, no. 423, pp. 782–792, 1993

    1993

  41. [41]

    The jackknife and the bootstrap for general stationary observations,

    H. R. K ¨unsch, “The jackknife and the bootstrap for general stationary observations,”The Annals of Statistics, vol. 17, no. 3, pp. 1217–1241, 1989

    1989

  42. [42]

    A nearest neighbor bootstrap for resampling hydrologic time series,

    U. Lall and A. Sharma, “A nearest neighbor bootstrap for resampling hydrologic time series,”Water Resources Research, vol. 32, no. 3, pp. 679–693, 1996

    1996

  43. [43]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European Conference on Computer Vision (ECCV), 2020

    2020

  44. [44]

    Layer Normalization

    J. L. Ba, J. R. Kiros, and G. E. Hinton, “Layer normalization,”arXiv preprint arXiv:1607.06450, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  45. [45]

    Gaussian Error Linear Units (GELUs)

    D. Hendrycks and K. Gimpel, “Gaussian error linear units (GELUs),” arXiv preprint arXiv:1606.08415, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  46. [46]

    Go- ing deeper with image transformers,

    H. Touvron, M. Cord, A. Sablayrolles, G. Synnaeve, and H. J ´egou, “Go- ing deeper with image transformers,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 32–42

    2021

  47. [47]

    FiLM: Visual reasoning with a general conditioning layer,

    E. Perez, F. Strub, H. de Vries, V . Dumoulin, and A. Courville, “FiLM: Visual reasoning with a general conditioning layer,” inProceedings of the AAAI Conference on Artificial Intelligence, 2018

    2018

  48. [48]

    Categorical reparameterization with Gumbel-Softmax,

    E. Jang, S. Gu, and B. Poole, “Categorical reparameterization with Gumbel-Softmax,” inInternational Conference on Learning Represen- tations (ICLR), 2017

    2017

  49. [49]

    Representation learning with contrastive predictive coding,

    A. van den Oord, Y . Li, and O. Vinyals, “Representation learning with contrastive predictive coding,”arXiv preprint, 2018

    2018

  50. [50]

    LoRA: Low-Rank Adaptation of Large Language Models

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, and W. Chen, “LoRA: Low-rank adaptation of large language models,” arXiv preprint arXiv:2106.09685, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  51. [51]

    A Time Series is Worth 64 Words: Long-term Forecasting with Transformers

    Y . Nie, N. H. Nguyen, P. Sinthong, and J. Kalagnanam, “A time series is worth 64 words: Long-term forecasting with transformers,” arXiv preprint arXiv:2211.14730, 2023. [Online]. Available: https: //arxiv.org/abs/2211.14730

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  52. [52]

    DAB-DETR: Dynamic anchor boxes are better queries for DETR,

    S. Liu, F. Li, H. Zhang, X. Yang, X. Qi, H. Su, J. Zhu, and L. Zhang, “DAB-DETR: Dynamic anchor boxes are better queries for DETR,” in International Conference on Learning Representations (ICLR), 2022

    2022

  53. [53]

    Mallat,A Wavelet Tour of Signal Processing: The Sparse Way, 3rd ed

    S. Mallat,A Wavelet Tour of Signal Processing: The Sparse Way, 3rd ed. Elsevier, 2008

    2008

  54. [54]

    XGBoost: A scalable tree boosting system,

    T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” inProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016, pp. 785–794

    2016

  55. [55]

    On a test of whether one of two random variables is stochastically larger than the other,

    H. B. Mann and D. R. Whitney, “On a test of whether one of two random variables is stochastically larger than the other,”The Annals of Mathematical Statistics, vol. 18, no. 1, pp. 50–60, 1947

    1947