pith. sign in

arxiv: 2606.06010 · v1 · pith:EKBWTCDUnew · submitted 2026-06-04 · 💻 cs.LG · cs.DB

Adaptive Oscillatory-State Alignment for Time Series Forecasting

Zhangyao Song , Ziqiong Li , Xiangfei Qiu , Chao Zha , Yinfei Xu , Tao Guo This is my paper

Pith reviewed 2026-06-28 02:48 UTC · model grok-4.3

classification 💻 cs.LG cs.DB
keywords time series forecastingoscillatory alignmentHilbert transformadaptive modelingnon-stationary signalsperiodic forecastinganalytic signal descriptors
0
0 comments X

The pith

AOSNET improves long-term forecasts by aligning local states to a learnable oscillatory prior instead of matching fixed templates.

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

The paper sets out to show that periodic patterns in time series are rarely fixed and instead vary through changes in amplitude, phase, and frequency. It replaces rigid template matching with a process that pulls analytic-signal descriptors from both the observed data and a global learnable prior, then uses those descriptors to decide how to blend or correct each local segment. A reader would care because this lets the model keep trustworthy observations while adjusting only where the prior and data disagree. If the approach holds, forecasts remain accurate even when the underlying oscillation drifts over long horizons.

Core claim

AOSNET reformulates periodic forecasting from fixed template matching to adaptive oscillatory-state alignment. It extracts analytic-signal descriptors from both the observed sequence and a learnable global oscillatory prior, then adaptively aligns local states through a descriptor-conditioned gate that selectively preserves reliable observations while softly correcting mismatched regions. The learned prior functions as a flexible oscillatory reference interpreted through local state dynamics rather than a rigid repeated template.

What carries the argument

The descriptor-conditioned gate that uses analytic-signal features to decide how much of the local observation versus the global prior to retain at each point.

If this is right

  • Accuracy stays competitive or superior across eight real benchmarks that contain varying degrees of non-stationarity.
  • Inference remains fast because the alignment uses a single learnable prior and a lightweight gate.
  • The performance margin widens as the controlled synthetic tests increase the strength of amplitude modulation, phase drift, or local frequency change.

Where Pith is reading between the lines

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

  • The same descriptor-plus-gate mechanism could be inserted into other sequence models that already handle long-range dependencies.
  • The learned prior might serve as an interpretable summary of the dominant oscillation even in datasets where periodicity is not labeled in advance.
  • If the gate learns to ignore the prior in highly non-oscillatory regions, the framework could naturally fall back to non-periodic modeling without extra switches.

Load-bearing premise

Oscillatory patterns in real time series change through amplitude modulation, phase drift, and local frequency shifts instead of staying rigidly periodic.

What would settle it

On synthetic series engineered with strong phase drift or frequency variation, if AOSNET shows no accuracy gain over fixed-template baselines, the advantage of adaptive alignment would be refuted.

Figures

Figures reproduced from arXiv: 2606.06010 by Chao Zha, Tao Guo, Xiangfei Qiu, Yinfei Xu, Zhangyao Song, Ziqiong Li.

Figure 1
Figure 1. Figure 1: Motivation of AOSNET. Existing periodic methods typically learn or select fixed periodic templates, which can be sensitive to period misspec￾ification, phase drift, and frequency variation. AOSNET instead represents temporal structure as adaptive local oscillatory states derived from amplitude, phase, and instantaneous-frequency descriptors. Amplitude modulation changes the signal strength without altering… view at source ↗
Figure 2
Figure 2. Figure 2: Overall architecture of AOSNET. The observed sequence and a learnable global oscillatory prior are mapped to analytic-signal descriptors, from which a descriptor-conditioned gate performs adaptive oscillatory-state alignment. The aligned representation is processed by two complementary forecasting paths: an attention path with channel-token attention, and a base path with direct temporal projection, fused … view at source ↗
Figure 3
Figure 3. Figure 3: Representative IaaS 96→96 forecasting visualization on a single channel. Each panel compares the same history window (gray), ground truth (black), and model forecast (colored) for AOSNET, TQNet, CycleNet, and iTransformer. AOSNET tracks the post-history state transition more faithfully, whereas the baselines either over-smooth the change point or drift toward rigid template-like forecasts. 2) Workload Fore… view at source ↗
Figure 4
Figure 4. Figure 4: Efficiency comparison on Electricity (look-back [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Controlled synthetic variants and their Hilbert-domain characterization. Each row shows a representative signal, its envelope, phase residual relative [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Stratified evaluation by non-stationarity severity. Test samples are binned into five quantiles (Q1=mildest, Q5=most severe). The [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Learned global oscillatory prior on Electricity. Variables are selected [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 10
Figure 10. Figure 10: Relation between prior oscillation strength and incoming channel [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Failure of fixed-template retrieval on the combined non-stationary dataset. Top-left: windows starting from the same cycle phase exhibit different [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Prediction visualization on four synthetic datasets (look-back [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Parameter sensitivity on four ETT datasets (horizon [PITH_FULL_IMAGE:figures/full_fig_p012_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Look-back length sensitivity on Electricity (horizon [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] view at source ↗
read the original abstract

Long-term time series forecasting benefits from inductive biases that expose recurring temporal structure. Existing periodic forecasting methods typically model recurrence through predefined periods, global spectral components, or fixed learnable templates. However, real-world temporal dynamics are rarely rigidly periodic: oscillatory behavior often evolves through amplitude modulation, phase drift, and local frequency variation. Under these conditions, fixed-template periodic modeling can become fundamentally mismatched to the underlying temporal states. We propose AOSNET, a Hilbert-guided forecasting framework that reformulates periodic forecasting from fixed template matching to adaptive oscillatory-state alignment. AOSNET extracts analytic-signal descriptors from both the observed sequence and a learnable global oscillatory prior, then adaptively aligns local states through a descriptor-conditioned gate that selectively preserves reliable observations while softly correcting mismatched regions. The learned prior serves not as a rigid repeated template but as a flexible oscillatory reference interpreted through local state dynamics. Experiments on eight benchmarks demonstrate state-of-the-art or highly competitive accuracy with fast inference speed. Controlled synthetic studies isolating amplitude modulation, phase drift, and local frequency variation confirm that the advantage of oscillatory-state alignment consistently increases as non-stationarity intensifies.

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 claims that existing periodic forecasting methods relying on predefined periods, global spectral components, or fixed learnable templates are mismatched to real-world time series exhibiting amplitude modulation, phase drift, and local frequency variation. It proposes AOSNET, a Hilbert-guided framework that extracts analytic-signal descriptors from both the observed sequence and a learnable global oscillatory prior, then applies a descriptor-conditioned gate to adaptively align local states by preserving reliable observations and softly correcting mismatches. The learned prior acts as a flexible reference rather than a rigid template. Experiments on eight benchmarks report state-of-the-art or competitive accuracy with fast inference, while controlled synthetic studies isolating the three non-stationarity types show that the advantage of oscillatory-state alignment grows with increasing non-stationarity.

Significance. If the central claims hold, the work offers a principled shift from fixed-template periodic modeling to adaptive alignment using analytic signals, which could improve long-term forecasting in domains with evolving oscillatory dynamics. The controlled synthetic experiments that isolate amplitude modulation, phase drift, and local frequency variation provide a clear falsifiable test of the method's robustness and are a methodological strength.

major comments (2)
  1. [Abstract/Method] Abstract and Method (central construction): the learnable global oscillatory prior is optimized on the same data used for alignment and forecasting; this creates a moderate circularity risk because it is unclear whether the descriptor-conditioned alignment metric supplies an independent reference or reduces to a fitted quantity by construction. An ablation separating prior optimization from the alignment objective is needed to substantiate the claim that the prior functions as a flexible oscillatory reference.
  2. [Experiments] Experiments section: the abstract asserts SOTA or competitive results on eight benchmarks, but without reported error bars, data splits, or baseline implementation details it is impossible to verify whether the gains are robust or attributable to the adaptive alignment rather than hyperparameter tuning or dataset-specific effects.
minor comments (2)
  1. [Abstract] Abstract: the phrase 'fast inference speed' is stated without quantitative comparison to baselines or complexity analysis; adding a brief runtime table or FLOPs count would strengthen the claim.
  2. [Method] Notation: the term 'analytic-signal descriptors' is introduced without an explicit equation or reference to the Hilbert transform definition in the main text; a short definition or pointer to the standard formula would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and positive assessment of the work's significance and the value of the controlled synthetic experiments. We address each major comment below and outline the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract/Method] Abstract and Method (central construction): the learnable global oscillatory prior is optimized on the same data used for alignment and forecasting; this creates a moderate circularity risk because it is unclear whether the descriptor-conditioned alignment metric supplies an independent reference or reduces to a fitted quantity by construction. An ablation separating prior optimization from the alignment objective is needed to substantiate the claim that the prior functions as a flexible oscillatory reference.

    Authors: We agree that joint optimization of the prior and the alignment objective introduces a potential circularity that merits explicit verification. While the prior is intended as a global reference and the descriptor-conditioned gate provides local adaptation, an ablation isolating the prior's optimization from the alignment loss would strengthen the claim. In the revision we will add such an ablation: one variant optimizes the prior solely via a reconstruction or forecasting loss without the alignment term, and another freezes the prior after an initial phase before applying alignment. This will clarify whether the adaptive alignment supplies an independent benefit beyond fitting the prior. revision: yes

  2. Referee: [Experiments] Experiments section: the abstract asserts SOTA or competitive results on eight benchmarks, but without reported error bars, data splits, or baseline implementation details it is impossible to verify whether the gains are robust or attributable to the adaptive alignment rather than hyperparameter tuning or dataset-specific effects.

    Authors: We acknowledge that the current experimental reporting lacks sufficient detail for full reproducibility and robustness assessment. In the revised manuscript we will report mean performance with standard deviations over at least three random seeds, explicitly state the train/validation/test splits (including any preprocessing or normalization), and provide additional implementation details for all baselines (hyperparameter search ranges, training protocols, and code references where applicable). These additions will allow readers to assess whether the reported gains are attributable to the proposed alignment mechanism. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes a standard neural architecture (AOSNET) with a learnable global oscillatory prior trained end-to-end on the forecasting data, followed by descriptor extraction and gated alignment. No equations or steps are shown that reduce a claimed prediction to a fitted quantity by construction, nor does the provided text invoke self-citations as load-bearing uniqueness theorems. The reformulation from fixed templates to adaptive alignment is presented as an architectural choice with external validation on benchmarks and controlled synthetics, making the derivation self-contained.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

Ledger entries are inferred directly from the abstract's description of the proposed components.

free parameters (1)
  • learnable global oscillatory prior
    A learnable reference signal introduced as the flexible oscillatory template; its parameters are fitted during training.
axioms (1)
  • domain assumption The Hilbert transform yields useful analytic-signal descriptors that capture amplitude and phase information suitable for state alignment.
    Invoked to extract descriptors from both the observed sequence and the learnable prior.
invented entities (1)
  • descriptor-conditioned gate no independent evidence
    purpose: To selectively preserve reliable observations while softly correcting mismatched regions during oscillatory-state alignment.
    New architectural component introduced to implement the adaptive alignment.

pith-pipeline@v0.9.1-grok · 5732 in / 1299 out tokens · 73985 ms · 2026-06-28T02:48:24.462022+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

41 extracted references · 5 canonical work pages · 1 internal anchor

  1. [1]

    Trans- formers in Time Series: A Survey,

    Q. Wen, T. Zhou, C. Zhang, W. Chen, Z. Ma, J. Yan, and L. Sun, “Trans- formers in Time Series: A Survey,”arXiv preprint arXiv:2202.07125, 2022

    work page arXiv 2022

  2. [2]

    TFB: Towards Comprehensive and Fair Benchmarking of Time Series Forecasting Methods,

    X. Qiu, J. Hu, L. Zhou, X. Wu, J. Du, B. Zhang, C. Guo, A. Zhou, C. S. Jensen, Z. Sheng, and B. Yang, “TFB: Towards Comprehensive and Fair Benchmarking of Time Series Forecasting Methods,” inInternational Conference on Very Large Data Bases (VLDB), Guangzhou, China, Aug. 26-30, 2024

    2024

  3. [3]

    Explor- ing Progress in Multivariate Time Series Forecasting: Comprehensive Benchmarking and Heterogeneity Analysis,

    Z. Shao, F. Wang, Y . Xu, W. Wei, C. Yu, Z. Zhang, D. Yao, T. Sun, G. Jin, X. Cao, G. Cong, C. S. Jensen, and X. Cheng, “Explor- ing Progress in Multivariate Time Series Forecasting: Comprehensive Benchmarking and Heterogeneity Analysis,”IEEE Transactions on Knowledge and Data Engineering, vol. 37, pp. 291–305, Jan. 2025

    2025

  4. [4]

    A Survey of Deep Learning for Time Series Forecasting: Taxonomy, Analysis and Future Directions,

    Q. Guo, B. Zhao, M. Song, and G. Zhong, “A Survey of Deep Learning for Time Series Forecasting: Taxonomy, Analysis and Future Directions,”IEEE Transactions on Knowledge and Data Engineering, pp. 1–20, 2026

    2026

  5. [5]

    Informer: Beyond Efficient Transformer for Long Sequence Time- Series Forecasting,

    H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, and W. Zhang, “Informer: Beyond Efficient Transformer for Long Sequence Time- Series Forecasting,” inProceedings of The AAAI Conference on Artificial Intelligence, Virtual Event, Feb. 2-9, 2021

    2021

  6. [6]

    Are Transformers Effective for Time Series Forecasting?

    A. Zeng, M. Chen, L. Zhang, and Q. Xu, “Are Transformers Effective for Time Series Forecasting?” inProceedings of The AAAI Conference on Artificial Intelligence, Washington, DC, USA, Feb. 7-14, 2023

    2023

  7. [7]

    Multi-Period Learning for Financial Time Series Forecasting,

    X. Zhang, Z. Huang, Y . Wu, X. Lu, E. Qi, Y . Chen, Z. Xue, Q. Wang, P. Wang, and W. Wang, “Multi-Period Learning for Financial Time Series Forecasting,” inProceedings of The 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Toronto, ON, Canada, Aug. 3-7, Jul. 2025, pp. 2848–2859. [Online]. Available: http://arxiv.org/abs/2511.08622

    work page arXiv 2025

  8. [8]

    FuXi: A Cascade Machine Learning Forecasting System for 15-Day Global Weather Forecast,

    L. Chen, X. Zhong, F. Zhang, Y . Cheng, Y . Xu, Y . Qi, and H. Li, “FuXi: A Cascade Machine Learning Forecasting System for 15-Day Global Weather Forecast,”npj Climate and Atmospheric Science, vol. 6, no. 1, p. 190, Nov. 2023

    2023

  9. [9]

    Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecast- ing,

    H. Wu, J. Xu, J. Wang, and M. Long, “Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecast- ing,” inAnnual Conference on Neural Information Processing Systems (NeurIPS), Virtual, Dec. 6-14, 2021

    2021

  10. [10]

    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,” in International Conference on Learning Representations (ICLR), Kigali, Rwanda, May 1-5, 2023

    2023

  11. [11]

    iTransformer: Inverted Transformers Are Effective for Time Series Forecasting,

    Y . Liu, T. Hu, H. Zhang, H. Wu, S. Wang, L. Ma, and M. Long, “iTransformer: Inverted Transformers Are Effective for Time Series Forecasting,” inInternational Conference on Learning Representations (ICLR), Vienna, Austria, May 7-11, 2024

    2024

  12. [12]

    Frequency Adaptive Normaliza- tion for Non-Stationary Time Series Forecasting,

    W. Ye, S. Deng, Q. Zou, and N. Gui, “Frequency Adaptive Normaliza- tion for Non-Stationary Time Series Forecasting,” inAnnual Conference on Neural Information Processing Systems (NeurIPS), Vancouver, BC, Canada, Dec. 10 - 15, 2024

    2024

  13. [13]

    TimeBridge: Non-Stationarity Matters for Long-Term Time Series Forecasting,

    P. Liu, B. Wu, Y . Hu, N. Li, T. Dai, J. Bao, and S.-T. Xia, “TimeBridge: Non-Stationarity Matters for Long-Term Time Series Forecasting,” in International Conference on Machine Learning (ICML), Vancouver, Canada, Jul. 13-19, 2025

    2025

  14. [14]

    Diffusion-based spatio-temporal channel prediction via non-stationarity decoupling,

    Z. Song, X. Zhang, L. Zhuang, T. Guo, X. Zhao, Y . Xu, and S. Jin, “Diffusion-based spatio-temporal channel prediction via non-stationarity decoupling,”IEEE Transactions on Cognitive Communications and Networking, vol. 12, pp. 7647–7661, 2026

    2026

  15. [15]

    CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Pat- terns,

    S. Lin, W. Lin, X. Hu, W. Wu, R. Mo, and H. Zhong, “CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Pat- terns,” inAnnual Conference on Neural Information Processing Systems (NeurIPS), Vancouver, Canada, Dec. 9-15, 2024

    2024

  16. [16]

    Temporal Query Network for Efficient Multivariate Time Series Forecasting,

    S. Lin, H. Chen, H. Wu, C. Qiu, and W. Lin, “Temporal Query Network for Efficient Multivariate Time Series Forecasting,” inInternational Conference on Machine Learning (ICML), ser. Proceedings of Machine Learning Research, vol. 267. BC, Canada, Jul. 13-19: PMLR / OpenReview.net, 2025. [Online]. Available: https://proceedings.mlr.press/v267/lin25e.html

    2025

  17. [17]

    FILM: Frequency Improved Legendre Memory Model for Long-Term Time Series Forecasting,

    T. Zhou, Z. Ma, Q. Wen, L. Sun, T. Yao, W. Yin, R. Jinet al., “FILM: Frequency Improved Legendre Memory Model for Long-Term Time Series Forecasting,” inAnnual Conference on Neural Information Processing Systems (NeurIPS), LA, USA, Nov. 28-Dec. 9, 2022

    2022

  18. [18]

    FreDF: Learning to Forecast in The Frequency Domain,

    H. Wang, L. Pan, Y . Shen, Z. Chen, D. Yang, Y . Yang, S. Zhang, X. Liu, H. Li, and D. Tao, “FreDF: Learning to Forecast in The Frequency Domain,” inInternational Conference on Learning Representations (ICLR), Singapore, Apr. 24-28, 2025. [Online]. Available: https://openreview.net/forum?id=4A9IdSa1ul

    2025

  19. [19]

    Amplifier: Bringing Attention to Neglected Low-Energy Components in Time Series Forecasting,

    J. Fei, K. Yi, W. Fan, Q. Zhang, and Z. Niu, “Amplifier: Bringing Attention to Neglected Low-Energy Components in Time Series Forecasting,” inProceedings of The AAAI Conference on Artificial Intelligence, Philadelphia, PA, USA, Feb. 25 - Mar. 4, 2025. [Online]. Available: https://doi.org/10.1609/aaai.v39i11.33267

    work page doi:10.1609/aaai.v39i11.33267 2025

  20. [20]

    TimeMixer: Decomposable Multiscale Mixing for Time Series Forecasting,

    S. Wang, H. Wu, X. Shi, T. Hu, H. Luo, L. Ma, J. Y . Zhang, and J. Zhou, “TimeMixer: Decomposable Multiscale Mixing for Time Series Forecasting,” inInternational Conference on Learning Representations (ICLR), Vienna, Austria, May 7-11, 2024

    2024

  21. [21]

    Revi- talizing Multivariate Time Series Forecasting: Learnable Decomposition with Inter-Series Dependencies and Intra-Series Variations Modeling,

    G. Yu, J. Zou, X. Hu, A. I. Aviles-Rivero, J. Qin, and S. Wang, “Revi- talizing Multivariate Time Series Forecasting: Learnable Decomposition with Inter-Series Dependencies and Intra-Series Variations Modeling,” in International Conference on Machine Learning (ICML), Vienna, Austria, Jul. 21-27, 2024

    2024

  22. [22]

    Parsimony or Capability? Decomposition Delivers Both in Long-Term Time Series Forecasting,

    J. Deng, F. Ye, D. Yin, X. Song, I. Tsang, and H. Xiong, “Parsimony or Capability? Decomposition Delivers Both in Long-Term Time Series Forecasting,” inAnnual Conference on Neural Information Processing Systems (NeurIPS), Vancouver, Canada, Dec. 9-15, 2024

    2024

  23. [23]

    Theory of Communication,

    D. Gabor, “Theory of Communication,”Journal of the Institution of Electrical Engineers, vol. 93, no. 26, pp. 429–457, 1946

    1946

  24. [24]

    Estimating and Interpreting the Instantaneous Frequency of a Signal. I. Fundamentals,

    B. Boashash, “Estimating and Interpreting the Instantaneous Frequency of a Signal. I. Fundamentals,”Proceedings of the IEEE, vol. 80, no. 4, pp. 520–568, 1992

    1992

  25. [25]

    FED- FOrmer: Frequency Enhanced Decomposed Transformer for Long-Term Series Forecasting,

    T. Zhou, Z. Ma, Q. Wen, X. Wang, L. Sun, and R. Jin, “FED- FOrmer: Frequency Enhanced Decomposed Transformer for Long-Term Series Forecasting,” inInternational Conference on Machine Learning (ICML),, Baltimore, MD, Jul. 17-23, 2022

    2022

  26. [26]

    MICN: Multi-Scale Local and Global Context Modeling for Long-Term Series Forecasting,

    H. Wang, J. Peng, F. Huang, J. Wang, J. Chen, and Y . Xiao, “MICN: Multi-Scale Local and Global Context Modeling for Long-Term Series Forecasting,” inInternational Conference on Learning Representations (ICLR), Kigali, Rwanda, May 1-5, 2023

    2023

  27. [27]

    SparseTSF: Modeling Long-Term Time Series Forecasting with *1k* Parameters,

    S. Lin, W. Lin, W. Wu, H. Chen, and J. Yang, “SparseTSF: Modeling Long-Term Time Series Forecasting with *1k* Parameters,” inInternational Conference on Machine Learning (ICML), ser. Proceedings of Machine Learning Research, vol. 235. Vienna, Austria, Jul. 21-27: PMLR / OpenReview.net, 2024, pp. 30 211–30 226. [Online]. Available: https://proceedings.mlr....

    2024

  28. [28]

    MoFo: Empowering Long-Term Time Series Forecasting with Periodic Pattern Modeling,

    J. Ma, B. Wang, Q. Huang, G. Wang, P. Wang, Z. Zhou, and Y . Wang, “MoFo: Empowering Long-Term Time Series Forecasting with Periodic Pattern Modeling,” inAnnual Conference on Neural Information Processing Systems (NeurIPS), San Diego, CA, USA, Dec. 2-7, 2026. [Online]. Available: https://openreview.net/forum?id=sbvLts2HqR

    2026

  29. [29]

    FITS: Modeling Time Series with10K Parameters,

    Z. Xu, A. Zeng, and Q. Xu, “FITS: Modeling Time Series with10K Parameters,” inInternational Conference on Learning Representations (ICLR), Vienna, Austria, May 7-11, 2024

    2024

  30. [30]

    Attention Is All You Need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention Is All You Need,” inAnnual Conference on Neural Information Processing Systems (NeurIPS), Long Beach, California, USA, Dec. 4-9, 2017

    2017

  31. [31]

    Channel, Trend and Periodic-Wise Representation Learning for Multivariate Long-Term Time Series Forecasting,

    Z. Song, N. Jiang, M. He, X. Zhao, and T. Guo, “Channel, Trend and Periodic-Wise Representation Learning for Multivariate Long-Term Time Series Forecasting,” inInternational Conference on Acoustics, Speech and Signal Processing (ICASSP). Barcelona, Spain, May 4- 8: IEEE, 2026, pp. 4821–4825

    2026

  32. [32]

    The Capacity and Robustness Trade-Off: Revisiting The Channel Independent Strategy for Multivariate Time Series Forecasting,

    L. Han, H.-J. Ye, and D.-C. Zhan, “The Capacity and Robustness Trade-Off: Revisiting The Channel Independent Strategy for Multivariate Time Series Forecasting,”IEEE Transactions on Knowledge and Data Engineering, 2024

    2024

  33. [33]

    Crossformer: Transformer Utilizing Cross- Dimension Dependency for Multivariate Time Series Forecasting,

    Y . Zhang and J. Yan, “Crossformer: Transformer Utilizing Cross- Dimension Dependency for Multivariate Time Series Forecasting,” inInternational Conference on Learning Representations (ICLR). Kigali, Rwanda, May 1-5: OpenReview.net, 2023. [Online]. Available: https://openreview.net/forum?id=vSVLM2j9eie

    2023

  34. [34]

    HDMixer: Hierarchical Dependency with Extendable Patch for Multivariate Time Series Forecasting,

    Q. Huang, L. Shen, R. Zhang, J. Cheng, S. Ding, Z. Zhou, and Y . Wang, “HDMixer: Hierarchical Dependency with Extendable Patch for Multivariate Time Series Forecasting,” inProceedings of The AAAI Conference on Artificial Intelligence, Vancouver, Canada, Feb. 20-27, 2024

    2024

  35. [35]

    SAMformer: Unlocking The Potential of Transformers in Time Series Forecasting with Sharpness-Aware Minimization and Channel-Wise Attention,

    R. Ilbert, A. Odonnat, V . Feofanov, A. Virmaux, G. Paolo, T. Palpanas, and I. Redko, “SAMformer: Unlocking The Potential of Transformers in Time Series Forecasting with Sharpness-Aware Minimization and Channel-Wise Attention,” inInternational Conference on Machine Learning (ICML), vol. 235. Vienna, Austria, Jul. 21-27: PMLR, 2024. [Online]. Available: ht...

    2024

  36. [36]

    The Fast Fourier Transform,

    E. O. Brigham and R. E. Morrow, “The Fast Fourier Transform,”IEEE Spectrum, 1967

    1967

  37. [37]

    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

  38. [38]

    Dropout: A Simple Way to Prevent Neural Networks from Overfitting,

    N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhut- dinov, “Dropout: A Simple Way to Prevent Neural Networks from Overfitting,”Journal of Machine Learning Research, vol. 15, no. 56, pp. 1929–1958, 2014

    1929

  39. [39]

    Pytorch: An Imperative Style, High-Performance Deep Learning Library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antigaet al., “Pytorch: An Imperative Style, High-Performance Deep Learning Library,” inAnnual Conference on Neural Information Processing Systems (NeurIPS), Van- couver, Canada, Dec. 8-14, 2019

    2019

  40. [40]

    Enhancing Time Series Forecasting through Selective Representation Spaces: A Patch Perspective,

    X. Wu, X. Qiu, H. Cheng, Z. Li, J. Hu, C. Guo, and B. Yang, “Enhancing Time Series Forecasting through Selective Representation Spaces: A Patch Perspective,” inAnnual Conference on Neural Information Pro- cessing Systems (NeurIPS), Vancouver, Canada, Dec. 9-14, 2025

    2025

  41. [41]

    Sparse-scale transformer with bidirectional awareness for time series forecasting,

    Y . Liu, B. Liu, S. Huang, G. Luo, W. Hu, M. Wang, and R. Hong, “Sparse-scale transformer with bidirectional awareness for time series forecasting,” inProceedings of The AAAI Conference on Artificial Intelligence. Singapore, January 20-27: AAAI Press, 2026, pp. 23 899– 23 907. [Online]. Available: https://doi.org/10.1609/aaai.v40i28.39566

    work page doi:10.1609/aaai.v40i28.39566 2026