Pith. sign in

REVIEW 5 major objections 8 minor 34 references

MTS-UNMixers: Multivariate Time Series Forecasting via Channel-Time Dual Unmixing

T0 review · 5 major / 8 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that treating history and future as one unmixed sequence with shared time bases and channel coefficients outperforms nine baselines in long-term multivariate forecasting.

desk verdict A reasonable Mamba-based extension of MTS-Mixers, but the headline accuracy claims are unverifiable until Table I is fixed and the direct predecessor baselines are added. read the letter →

arxiv 2411.17770 v1 pith:CV654NX6 submitted 2024-11-26 cs.LG

Xuanbing Zhu , Dunbin Shen , Zhongwen Rao , Huiyi Ma , Yingguang Hao , Hongyu Wang This is my paper · ORCID
classification cs.LG
keywords multivariatetimeseriesforecastingunmixingMambabidirectionalstatespacemodelssharedcomponentslong-termchannel-timedecomposition
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Multivariate time series are hard to forecast because trends, cycles, and inter-variable correlations arrive mixed together, and because standard models map history to future without an explicit account of which parts carry over. This paper argues that both problems are solved by decomposing the entire series along two axes at once: time is a mixture of a few shared trend and cycle basis signals, and channels are a mixture of a few shared correlation bases that persist across the whole series. The proposed MTS-UNMixers learns these shared components with a causal Mamba encoder for time and a bidirectional Mamba encoder for channels, then reconstructs the past and predicts the future by recombining the same parts. On the ETT, Weather, Traffic, and Electricity benchmarks it reports top-two performance in every scenario and first place in 56 cases, with ablations showing the time unmixing module contributes the largest gain.

What carries the argument

The carrying object is the channel-time dual unmixing factorization, a pair of matrix decompositions applied to the same observed series. Along time, the data are written as $X = A_t S_t$ with $A_t$ a small set of trend and cycle basis signals; along channels, the data are written as $X = A_c S_c$ with $S_c$ a small set of tick-wise correlation bases. The key move is that $A_t$ and $S_c$ are shared between the historical and future windows, while the coefficient matrix $S_t$ and the channel basis $A_c$ are allowed to differ, so forecasting reduces to recombining components that were already extracted. A vanilla Mamba block, a linear-time selective state space model, estimates the time-dependent coefficients because time has a causal direction, and a bidirectional Mamba block estimates the channel-correlated bases because channel relations are noncausal; softmax enforces the two constraints on the coefficients, and an L1 loss trains reconstruction and prediction together.

What would settle it

Train MTS-UNMixers on an early segment of a dataset and evaluate on a later segment whose channel correlations are known to shift, such as electricity demand before and after a structural change; if a version with separate history and future components matches or beats the shared-component version, then the shared-component invariance, not the unmixing itself, is carrying the reported gains.

Watch

Extended reading notes

Core claim

The central claim is that the historical and future windows are two views of one latent mixture, so the same temporal basis matrix $A_t$ and the same channel coefficient matrix $S_c$ can carry both reconstruction and prediction. Writing the observed series as $X = A_c S_c$ along channels and $X = A_t S_t$ along time, the future is produced as $\hat{X} = A_t \hat{S}_t$ and $\hat{X} = \hat{A}_c S_c$, with the shared components enforcing continuity of trends and of inter-variable relationships. The coefficient matrices are constrained to be non-negative and to sum to one, which the authors present as the source of physical interpretability. The paper's experiments compare this design against nine baselines on seven datasets at horizons 96, 192, 336, and 720, and report that the full model ranks first or second everywhere, that removing the time unmixing module causes the largest error increase, and that the bidirectional channel encoder matters most on the weather data.

Load-bearing premise

The load-bearing premise is that the historical and future windows share the same temporal basis signals and the same channel coefficient matrix, so the components learned from the past are still the right components for the future.

Editorial extensions

If this is right

  • Forecast horizons can be extended by recombining the same shared bases with new coefficients, so the model does not need to learn a separate decoder for each prediction length.
  • The shared components give a concrete inspection point: the learned trend and cycle bases and the per-channel loadings are directly readable as the model's explanation of what carries over from past to future.
  • Because the forecast is built from components observed in history, the model is structurally prevented from inventing channel relationships that never appeared in the training window.
  • The ablation results point to time unmixing as the dominant source of accuracy, so improvements to the temporal basis estimation should yield larger gains than further channel modeling.
  • If the shared-component design generalizes, the same reconstruction-and-prediction decoder can be applied to any horizon on all four benchmark families without retraining the encoder.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the same unmix-and-recombine decoder could be applied to other sequence tasks such as missing-value imputation and anomaly detection, because those tasks also need a shared representation that can regenerate an observed window.
  • Editorial inference: if the shared temporal basis is truly invariant, forecasting error should increase smoothly as the forecast window moves further from the training region, and a reader could test this by using the trained $A_t$ to decode windows far beyond the longest horizon evaluated.
  • Editorial inference: the strict sharing of $S_c$ is a strong prior, and a natural stress test is to allow the channel coefficient matrix to drift slowly with time and check whether accuracy improves on datasets whose channel correlations change between training and test periods.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 8 minor

Summary. The paper proposes MTS-UNMixers, a multivariate time series forecasting architecture that decomposes the input into temporal basis signals and channel coefficient matrices, and also into channel basis signals and temporal coefficient matrices. The historical and future windows are treated as a unified sequence that shares the temporal basis matrix At and the channel coefficient matrix Sc. A vanilla Mamba network estimates the time-dependent coefficients, while a bidirectional Mamba network estimates the channel-correlated bases. The authors claim that this dual unmixing plus sharing mechanism provides an explicit, physically interpretable mapping between history and future, and that it significantly outperforms existing methods on seven benchmark datasets (ETTh1, ETTh2, ETTm1, ETTm2, Weather, Traffic, Electricity) across four prediction horizons.

Significance. If the empirical claims were fully supported, the paper would make a useful contribution: it combines a relatively interpretable matrix-factorization view of time series with modern Mamba-based sequence models, and it includes an ablation study, a lookback-window analysis, and an efficiency comparison. The code is also made publicly available. However, the central claim of state-of-the-art or near-state-of-the-art performance currently rests on a single comparison table whose internal numbers are inconsistent, and the table omits the most directly relevant baselines. The shared-basis stationarity assumption is also a genuine conceptual risk, but it is secondary to the empirical verification problem. With corrected and completed experiments, the approach could merit publication; in its current form the evidence is not sufficient to support the headline claims.

major comments (5)
  1. [§V-B-b, Table I] The '1st Count' row in Table I is internally inconsistent. The table has 7 datasets x 4 horizons x 2 metrics = 56 cells, and the text states that MTS-UNMixers 'ranked first in 56 cases.' Yet the printed row gives 23 and 33 for MTS-UNMixers (which sums to 56 if read per metric), while also crediting TimeXer with 9 and 3, TimeMixer with 3, and DLinear with 1, for a total of 72 claimed first places. If the row is instead read as per-model totals, the sum is 23+33+9+3+3+1 = 72. Either way, the count cannot be correct, and the reader cannot determine which model actually wins which cells. This makes the headline performance advantage unverifiable.
  2. [§V-B-a, Table I] The comparison omits MTS-Mixers [12], the direct predecessor that this work explicitly extends, and iTransformer [11], a strong transformer baseline cited in the introduction. The 'Comparison Methods' paragraph says nine models but then enumerates ten (PatchTST, FEDformer, Autoformer, Stationary Transformer, TimeXer, DLinear, FITS, TiDE, TimesNet, TimeMixer). Without MTS-Mixers and iTransformer, the claim that MTS-UNMixers 'significantly outperform[s] existing methods' is not supported against the most relevant alternatives, and the discrepancy between 'nine' and ten listed models further undermines confidence in the comparison.
  3. [§V-B, Table I and §V-C, Table II] No error bars, number of seeds, or statistical significance tests are reported anywhere in the experimental section. The word 'significantly' in the Abstract and Section V-B-b is not backed by any variance estimate or test. Please report at least mean and standard deviation over multiple seeds for the key comparisons, and, where possible, paired significance tests for the differences against the strongest baselines.
  4. [§IV-C, Eqs. (7)-(9)] The sharing of At and Sc across historical and future windows is imposed by construction, not validated. If the latent mixing structure drifts between the observed and forecast windows, the shared-component constraint becomes a bias rather than a benefit. Because the network learns all factors from data, this sharing is an inductive bias, not a verified property of the data. Please add an experiment that relaxes the sharing (for example, separate basis/coefficient matrices for reconstruction and prediction) or otherwise measures the drift of the learned bases and coefficients across windows.
  5. [§III-B-c and §IV-D] The optimization problem in Eq. (11) imposes sum-to-one and non-negativity constraints on the coefficient matrices, but the network implementation only applies softmax to Sc, St, and Sp, while At, Ac, and Ap remain unconstrained, and the loss in Eq. (21) contains no constraint penalties. Please clarify how the constraints from Eqs. (3)-(6) are enforced in the implementation, or revise the formulation, because the paper's 'physical interpretability' claim depends on these constraints.
minor comments (8)
  1. [§V heading] The section heading reads 'EXPERIMENT RESLUTS' and should be 'EXPERIMENT RESULTS'.
  2. [§V-B-a] The sentence 'we compared the proposed MTS-UNMixers with nine well-established and advanced models' is followed by a list of ten models; please correct the count.
  3. [§V-E] The model efficiency paragraph says the ranking is based on 'Informer, Autoformer, FEDformer, PatchTST, TimesNet, TimeXer, and our proposed MTS-UNMixers, totaling six models,' but seven models are listed.
  4. [§IV-B-b] In the bidirectional Mamba equations, the symbol X is used both for the original historical sequence and for its linearly transformed version; please use distinct notation to avoid ambiguity.
  5. [Fig. 2] Figure 2 contains a '???????' placeholder near the top, which should be replaced with the actual label or removed before publication.
  6. [Fig. 6] The captions contain typos: 'Infomer' should be 'Informer' and 'Transfomer' should be 'Transformer'.
  7. [§V-D] The phrase 'visualized in Figure 3' in the model efficiency analysis should refer to Figure 8, not Figure 3.
  8. [Table I] The rows labeled 'A VG.' should be 'Avg.' for consistency with the other rows.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the sharing constraints are explicit modeling assumptions tested empirically, not fitted predictions passed off as derivations, and the self-citation to MTS-Mixers is not load-bearing.

full rationale

The paper makes no claim to derive its forecasts from first principles; it proposes an architecture and validates it on benchmarks. The shared temporal basis A_t and shared channel coefficient matrix S_c in Eqs. (7)-(9) are explicitly introduced as an 'explicit mapping' assumption, i.e., an inductive bias learned from data, not a quantity fitted to the target and then renamed a prediction. The prediction heads in Eqs. (17)-(18) are supervised by the L1 loss (Eq. 21) against held-out future values, so the empirical comparisons in Table I are genuine out-of-sample evaluations. The only self-citation, to MTS-Mixers [12], appears in the related-work discussion and is not used to justify any step of the method; no uniqueness theorem or external authority is invoked. The 'physical interpretability' language is supported only by the sum-to-one and non-negativity constraints, which is a definitional design choice rather than a circular reduction of the forecasting claim. Concerns about Table I's count consistency and the omission of the MTS-Mixers baseline are correctness or empirical-validity issues, not circularity. Accordingly, the derivation chain is self-contained with respect to the circularity criteria.

Assumptions & free parameters 5 free parameters · 4 assumptions · 2 invented entities

The paper's central accuracy claim rests mainly on fitting a neural network, so many learned parameters are free. The notable assumptions are the stationarity of the shared bases and coefficients across the prediction horizon and the faithfulness of the softmax-based unmixing to the stated nonnegative sum-to-one model. The 'interpretable' bases have no independent evidence and function as latent variables.

free parameters (5)
  • learnable basis matrices A_c, A_p, A_t
    The channel reconstruction basis, channel prediction basis, and temporal basis are learned from the training data (Eqs. 17-18); their values are not reported.
  • learnable coefficient matrices S_t, S_p
    Temporal reconstruction and prediction coefficients are learned and softmax-normalized (Section IV-C-b); not reported.
  • Mamba/Bi-Mamba parameters
    State-space matrices A, B, C, convolutions, and linear layers in the vanilla and bidirectional Mamba blocks are learned (Section IV-B).
  • projection layer weights
    The final linear layer that fuses temporal and channel predictions (Eq. 19).
  • hyper-parameters k1, k2, patch size, depth, hidden size, lambda1, lambda2, learning rate, batch size, epochs
    Not stated in the paper; the number of bases, patch size, and loss weights are chosen by hand and not reported.
assumptions (4)
  • domain assumption Historical and future windows share the same temporal basis matrix A_t and channel coefficient matrix S_c (Eqs. 7-9).
    The explicit-mapping mechanism assumes this stationarity; if the mixing structure drifts, the shared components hurt prediction.
  • domain assumption The time series can be represented as nonnegative, sum-to-one mixtures of a small number of bases.
    Eqs. 3-4 impose sum-to-one and nonnegativity; the network only enforces these via softmax on coefficients, not on bases.
  • domain assumption Mamba's selective state-space model is a suitable estimator of temporal coefficients, and bidirectional Mamba captures channel correlations.
    No theoretical or empirical justification beyond benchmark performance; this is the design premise of the encoder.
  • standard math Standard matrix multiplication and the L1 norm measure meaningful reconstruction error.
    Elementary linear algebra used in Eqs. 10-11.
invented entities (2)
  • time basis signals {a_k}
    purpose: Represent the supposed interpretable trends and cycles that are shared across history and future.
    No experiment shows these learned bases correspond to actual physical trends or cycles; they are latent factors without external validation.
  • channel correlation bases
    purpose: Represent the supposed common inter-variable patterns shared across all channels.
    The paper claims physical interpretability but reports no interpretability metric or qualitative validation of the learned bases.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MTS-UNMixers: Multivariate Time Series Forecasting via Channel-Time Dual Unmixing." pith.science (2026). https://pith.science/paper/CV654NX6

@misc{pith2026241117770,
  author       = {Pith},
  title        = {Pith review of: MTS-UNMixers: Multivariate Time Series Forecasting via Channel-Time Dual Unmixing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CV654NX6}},
  note         = {Machine review of arXiv:2411.17770}
}
read the original abstract

Multivariate time series data provide a robust framework for future predictions by leveraging information across multiple dimensions, ensuring broad applicability in practical scenarios. However, their high dimensionality and mixing patterns pose significant challenges in establishing an interpretable and explicit mapping between historical and future series, as well as extracting long-range feature dependencies. To address these challenges, we propose a channel-time dual unmixing network for multivariate time series forecasting (named MTS-UNMixer), which decomposes the entire series into critical bases and coefficients across both the time and channel dimensions. This approach establishes a robust sharing mechanism between historical and future series, enabling accurate representation and enhancing physical interpretability. Specifically, MTS-UNMixers represent sequences over time as a mixture of multiple trends and cycles, with the time-correlated representation coefficients shared across both historical and future time periods. In contrast, sequence over channels can be decomposed into multiple tick-wise bases, which characterize the channel correlations and are shared across the whole series. To estimate the shared time-dependent coefficients, a vanilla Mamba network is employed, leveraging its alignment with directional causality. Conversely, a bidirectional Mamba network is utilized to model the shared channel-correlated bases, accommodating noncausal relationships. Experimental results show that MTS-UNMixers significantly outperform existing methods on multiple benchmark datasets. The code is available at https://github.com/ZHU-0108/MTS-UNMixers.

Figures

Figures reproduced from arXiv: 2411.17770 by the authors.

Figure 1
Figure 1. Mixing problem (taking weather data as an example). (a) shows the original plot of the seven variables and their [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The framework of MTS-UNMixers comprises two main components: temporal unmixing and channel unmixing. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Structure of the Mamba Block. B. Unmixing Encoder a) Temporal (Channel-wise Representation) Unmixing Encoder: The Time Encoder first processes the historical sequence through a blocking operation, resulting in Xp ∈ R N×Tp×P , where the sequence is divided into small segments (tokens) for feature extraction. Subsequently, these tokens are processed by the Mamba block, which leverages a SSM to capture complex temporal… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Structure of the Bi-Mamba Block. C. Unmixing Decoder The unmixing decoder reconstructs the historical sequence and predicts the future sequence using the extracted features, followed by a projection layer to generate the final model output. a) Channel Reconstruction an…
Figure 5
Figure 5. Figure 5: Reconstruction visualization of MTS-UNMixers under different configurations on ETTh1, Weather, and Electricity [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Visualization of MTS-UNMixers’ prediction results on Electricity. The input length [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: The predictive performance of different models for [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: These four graphs show the speed and parameter efficiency of the model at different historical and future sequence [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 26 canonical work pages

  1. [12]

    MTS-Mixers: Multivariate Time Series Forecasting via Factorized Temporal and Channel Mix- ing,

    Z. Li, Z. Rao, L. Pan, and Z. Xu, “MTS-Mixers: Multivariate Time Series Forecasting via Factorized Temporal and Channel Mix- ing,” arXiv preprint arXiv:2302.04501, 2023. [Online]. Available: https://arxiv.org/abs/2302.04501

    arXiv 2023

  2. [11]

    Y . Liu, T. Hu, H. Zhang, H. Wu, S. Wang, L. Ma, and M. Long, ”iTransformer: Inverted Transformers Are Effective for Time Series Forecasting,” arXiv preprint arXiv:2310.06625 , 2023

    arXiv 2023

  3. [1]

    ”Bayesian forecasting for financial risk management, pre and post the global financial crisis.” Journal of Forecasting, vol

    Chen, Cathy WS, Gerlach, Richard, Lin, Edward MH, and Lee, WCW. ”Bayesian forecasting for financial risk management, pre and post the global financial crisis.” Journal of Forecasting, vol. 31, no. 8, pp. 661– 687, 2012

    work page 2012

  4. [2]

    ”Multi- variate time series dataset for space weather data analytics.” Scientific Data, vol

    Angryk, Rafal A, Martens, Petrus C, Aydin, Berkay, Kempton, Dustin, Mahajan, Sushant S, Basodi, Sunitha, Ahmadzadeh, Azim, Cai, Xumin, Filali Boubrahimi, Soukaina, Hamdi, Shah Muhammad, et al. ”Multi- variate time series dataset for space weather data analytics.” Scientific Data, vol. 7, no. 1, pp. 227, 2020

    work page 2020

  5. [3]

    ”Can deep learning beat numerical weather prediction?” Philosophical Transactions of the Royal Society A , vol

    Schultz, Martin G, Betancourt, Clara, Gong, Bing, Kleinert, Felix, Langguth, Michael, Leufen, Lukas Hubert, Mozaffari, Amirpasha, and Stadtler, Scarlet. ”Can deep learning beat numerical weather prediction?” Philosophical Transactions of the Royal Society A , vol. 379, no. 2194, pp. 20200097, 2021

    work page 2021

  6. [4]

    ”Towards efficient electricity forecasting in residential and commercial buildings: A novel hybrid CNN with a LSTM-AE based framework.” Sensors, vol

    Khan, Zulfiqar Ahmad, Hussain, Tanveer, Ullah, Amin, Rho, Seungmin, Lee, Miyoung, and Baik, Sung Wook. ”Towards efficient electricity forecasting in residential and commercial buildings: A novel hybrid CNN with a LSTM-AE based framework.” Sensors, vol. 20, no. 5, pp. 1399, 2020

    work page 2020

  7. [5]

    ”Energy forecasting with robust, flexible, and explainable machine learning algorithms.” AI Magazine, vol

    Zhu, Zhaoyang, Chen, Weiqi, Xia, Rui, Zhou, Tian, Niu, Peisong, Peng, Bingqing, Wang, Wenwei, Liu, Hengbo, Ma, Ziqing, Gu, Xinyue, et al. ”Energy forecasting with robust, flexible, and explainable machine learning algorithms.” AI Magazine, vol. 44, no. 4, pp. 377–393, 2023

    work page 2023

  8. [6]

    ”Freeway performance measurement system: mining loop detector data.” Transportation Research Record , vol

    Chen, Chao, Petty, Karl, Skabardonis, Alexander, Varaiya, Pravin, and Jia, Zhanfeng. ”Freeway performance measurement system: mining loop detector data.” Transportation Research Record , vol. 1748, no. 1, pp. 96–102, 2001

    work page 2001

Show all 34 references
  1. [7]

    ”Towards spatio-temporal aware traffic time series forecast- ing.” In 2022 IEEE 38th International Conference on Data Engineering (ICDE), pp

    Cirstea, Razvan-Gabriel, Yang, Bin, Guo, Chenjuan, Kieu, Tung, and Pan, Shirui. ”Towards spatio-temporal aware traffic time series forecast- ing.” In 2022 IEEE 38th International Conference on Data Engineering (ICDE), pp. 2900–2913, IEEE, 2022

    2022

  2. [8]

    ”Enabling time-dependent uncertain eco-weights for road networks.” In Proceedings of Workshop on Managing and Mining Enriched Geo-Spatial Data , pp

    Ma, Yu, Yang, Bin, and Jensen, Christian S. ”Enabling time-dependent uncertain eco-weights for road networks.” In Proceedings of Workshop on Managing and Mining Enriched Geo-Spatial Data , pp. 1–6, 2014

    2014

  3. [9]

    P. Chen, Y . Zhang, Y . Cheng, Y . Shu, Y . Wang, Q. Wen, B. Yang, and C. Guo, ”Pathformer: Multi-scale Transformers with Adaptive Pathways for Time Series Forecasting,” International Conference on Learning Representations (ICLR), 2024

    2024

  4. [10]

    TimeMixer: Decompos- able Multiscale Mixing for Time Series Forecasting

    Wang, Shiyu, Wu, Haixu, Shi, Xiaoming, Hu, Tengge, Luo, Huakun, Ma, Lintao, Zhang, James Y ., and Zhou, Jun. TimeMixer: Decompos- able Multiscale Mixing for Time Series Forecasting. In International Conference on Learning Representations (ICLR) , 2024

    2024

  5. [13]

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

    Nie, Yuqi, Nguyen, Nam H., Sinthong, Phanwadee, and Kalagnanam, Jayant. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers. In The Eleventh International Conference on Learning Representations

  6. [14]

    FEDformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting,

    T. Zhou, Z. Ma, Q. Wen, X. Wang, L. Sun, and R. Jin, “FEDformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting,” in *Proceedings of the 39th International Conference on Machine Learning (ICML)*, Baltimore, Maryland, 2022

    2022

  7. [15]

    Auto- former: Decomposition transformers with auto-correlation for long-term series forecasting

    Wu, Haixu, Xu, Jiehui, Wang, Jianmin, and Long, Mingsheng. Auto- former: Decomposition transformers with auto-correlation for long-term series forecasting. Advances in Neural Information Processing Systems , 34:22419–22430, 2021

    2021

  8. [16]

    Non- stationary transformers: Exploring the stationarity in time series fore- casting

    Liu, Yong, Wu, Haixu, Wang, Jianmin, and Long, Mingsheng. Non- stationary transformers: Exploring the stationarity in time series fore- casting. Advances in Neural Information Processing Systems , 35:9881– 9893, 2022

    2022

  9. [17]

    TSMixer: Lightweight MLP-Mixer Model for Multivariate Time Series Forecasting,

    V . Ekambaram, A. Jati, N. Nguyen, P. Sinthong, and J. Kalagnanam, “TSMixer: Lightweight MLP-Mixer Model for Multivariate Time Series Forecasting,” in *Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining*, 2023, pp. 459-469

    2023

  10. [18]

    Are transformers effective for time series forecasting? In Proceedings of the AAAI Conference on Artificial Intelligence , volume 37, number 9, pages 11121–11128, 2023

    Zeng, Ailing, Chen, Muxi, Zhang, Lei, and Xu, Qiang. Are transformers effective for time series forecasting? In Proceedings of the AAAI Conference on Artificial Intelligence , volume 37, number 9, pages 11121–11128, 2023

    2023

  11. [19]

    Liu, M., Zeng, A., Chen, M., Xu, Z., Lai, Q., Ma, L., & Xu, Q. (2022). Scinet: Time series modeling and forecasting with sample convolution and interaction. Advances in Neural Information Processing Systems , 35, 5816–5828

    2022

  12. [20]

    TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis

    Wu, Haixu, Hu, Tengge, Liu, Yong, Zhou, Hang, Wang, Jianmin, and Long, Mingsheng. TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis. In The Eleventh International Conference on Learning Representations

  13. [21]

    Luo, D., & Wang, X. (2024). Moderntcn: A modern pure convolution structure for general time series analysis. In The Twelfth International Conference on Learning Representations

    2024

  14. [22]

    Salinas, D., Flunkert, V ., Gasthaus, J., & Januschowski, T. (2020). DeepAR: Probabilistic forecasting with autoregressive recurrent net- works. International Journal of Forecasting, 36(3), 1181–1191. Elsevier

    2020

  15. [23]

    Zhao, Z., Chen, W., Wu, X., Chen, P. C. Y ., & Liu, J. (2017). LSTM network: A deep learning approach for short-term traffic forecast. IET Intelligent Transport Systems, 11(2), 68–75. Wiley Online Library

    2017

  16. [24]

    Gu and T

    A. Gu and T. Dao, ”Mamba: Linear-time sequence modeling with selective state spaces,” arXiv preprint arXiv:2312.00752 , 2023

    2023 arXiv

  17. [25]

    FMamba: Mamba based on Fast-attention for Multivariate Time-series Forecast- ing,

    S. Ma, Y . Kang, P. Bai, and Y .-B. Zhao, “FMamba: Mamba based on Fast-attention for Multivariate Time-series Forecast- ing,” arXiv preprint arXiv:2407.14814, 2024. [Online]. Available: https://arxiv.org/abs/2407.14814

    2024 arXiv

  18. [26]

    SiMBA: Simplified Mamba- Based Architecture for Vision and Multivariate Time series,

    B. N. Patro and V . S. Agneeswaran, “SiMBA: Simplified Mamba- Based Architecture for Vision and Multivariate Time series,” arXiv preprint arXiv:2403.15360, 2024. [Online]. Available: https://arxiv.org/abs/2403.15360

    2024 arXiv

  19. [27]

    VMRNN: Integrating Vision Mamba and LSTM for Efficient and Accurate Spatiotemporal Forecasting,

    Y . Tang, P. Dong, Z. Tang, X. Chu, and J. Liang, “VMRNN: Integrating Vision Mamba and LSTM for Efficient and Accurate Spatiotemporal Forecasting,” in *Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR) Workshops*, Jun. 2024, pp. 5663-5673

    2024

  20. [28]

    M. A. Ahamed and Q. Cheng, ”TimeMachine: A Time Series is Worth 4 Mambas for Long-term Forecasting,” arXiv preprint arXiv:2403.09898 , 2024

    2024 arXiv

  21. [29]

    Yuxuan Wang, Haixu Wu, Jiaxiang Dong, Guo Qin, Haoran Zhang, Yong Liu, Yunzhong Qiu, Jianmin Wang, Mingsheng Long, Timexer: Empowering transformers for time series forecasting with exogenous variables, arXiv preprint arXiv:2402.19072, 2024

    2024 arXiv

  22. [30]

    Efficiently Modeling Long Sequences with Structured State Spaces,

    A. Gu, K. Goel, and C. R ´e, “Efficiently Modeling Long Sequences with Structured State Spaces,” in The International Conference on Learning Representations (ICLR), 2022

    2022

  23. [31]

    Deep state space models for time series forecasting,

    S. S. Rangapuram, M. W. Seeger, J. Gasthaus, L. Stella, Y . Wang, and T. Januschowski, “Deep state space models for time series forecasting,” Advances in Neural Information Processing Systems, vol. 31, 2018

    2018

  24. [32]

    SSDNet: State Space Decomposition Neural Network for Time Series Forecasting,

    Y . Lin, I. Koprinska, and M. Rana, “SSDNet: State Space Decomposition Neural Network for Time Series Forecasting,” in 2021 IEEE Interna- tional Conference on Data Mining (ICDM) , 2021, pp. 370-378, doi: 10.1109/ICDM51629.2021.00048

    2021

  25. [33]

    FITS: Modeling Time Series with 10k Parameters

    Xu, Zhijian, Zeng, Ailing, and Xu, Qiang. FITS: Modeling Time Series with 10k Parameters. In The Twelfth International Conference on Learning Representations

  26. [34]

    Long-term Forecasting with TiDE: Time- series Dense Encoder

    Das, Abhimanyu, Kong, Weihao, Leach, Andrew, Mathur, Shaan K., Sen, Rajat, and Yu, Rose. Long-term Forecasting with TiDE: Time- series Dense Encoder. Transactions on Machine Learning Research

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.