Are Data Embeddings effective in time series forecasting?

Anil Pahwa; Balasubramaniam Natarajan; Reza Nematirad

REVIEW 3 major objections 5 minor 39 references

Are Data Embeddings effective in time series forecasting?

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Removing embedding layers generally improves time series forecasts.

desk verdict First broad ablation of embedding layers in time series models, with a plausible directional finding, but the no-embedding condition is confounded by ad hoc preprocessing and the statistical evidence is shaky. read the letter →

arxiv 2505.20716 v1 pith:XRA7MSPJ submitted 2025-05-27 cs.LG

Reza Nematirad , Anil Pahwa , Balasubramaniam Natarajan This is my paper · ORCID

classification cs.LG

keywords timeseriesforecastingdataembeddingablationstudymodelsimplificationmultivariateaccuracycomputationalefficiencylayers

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

This paper asks whether the embedding layers common in modern time-series forecasting models actually pay for themselves. Across fifteen state-of-the-art models and four benchmark datasets, the authors ablate the embedding components and find that removing them does not hurt accuracy in the vast majority of configurations, and often improves MSE and MAE while reducing training time and memory use. The paper's central point is that these gains frequently exceed the tiny differences, a few thousandths in error metrics, that separate competing state-of-the-art models. If the finding holds, it implies that a large share of recent architectural complexity in forecasting is unnecessary, and that simpler embedding-free variants deserve to be standard baselines. The authors are careful to say this does not prove embeddings are never useful, only that their widespread inclusion should be justified empirically.

What carries the argument

The load-bearing mechanism is the embedding-layer ablation itself: for each of the fifteen models, the authors identify which of the five embedding families, value, temporal, positional, inverted, or patch, is used, bypass that layer, and reconcile the input dimensions with the downstream model through permutation and concatenation of the raw input. Training and evaluation are otherwise kept identical, so any change in MSE, MAE, training time, or memory is attributed to the presence or absence of the embedding layer. The classification of embedding techniques supplies the vocabulary for deciding what counts as an embedding in each architecture.

What would settle it

Run the same fifteen-model ablation but, instead of permutation and concatenation, substitute a fixed random linear projection with the embedding's output dimension, so the downstream network sees identically shaped input with no learned embedding. If accuracy no longer improves (or degrades) relative to the original models, the paper's attribution to embedding removal would be undermined. A second check would be to test on a dataset with irregular sampling or a longer input window and find configurations where embedding-free variants consistently lose.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that data embedding layers, including value, temporal, positional, inverted, and patch embeddings, are often redundant in multivariate time-series forecasting. After reproducing fifteen published models and rerunning each with its embedding layer bypassed, the paper reports that over 95% of model-horizon configurations improve in MSE or MAE without the embedding, with average error reductions around 0.02 to 0.03 on the ETT datasets. In several cases, such as ETSformer on ETTh1 at horizon 720, the improvement is large, with MSE down by 0.360. Removing embeddings also cuts training time and memory in most configurations; a minority of models, such as EDformer, become slower or more memory-hungry because the replacement preprocessing creates extra tensors and misaligns with GPU kernel tile sizes. The paper concludes that raw multivariate inputs often carry enough information on their own, and recommends that the community assess embeddings critically before adding complexity.

Load-bearing premise

The load-bearing premise is that replacing an embedding layer with permutation and concatenation of the raw input is equivalent to simply deleting the embedding; if that reshaping itself changes what information reaches the model, the measured gains cannot be cleanly attributed to removing embeddings.

Editorial extensions

If this is right

Embedding-free variants of popular forecasting models can serve as stronger, cheaper baselines than the original releases.
Reported differences between state-of-the-art models of a few thousandths in MSE or MAE are smaller than the typical gain from removing embeddings, so model rankings may be less meaningful than architecture simplifications.
Longer forecasting horizons tend to benefit more from embedding removal, suggesting simplified designs are especially relevant for long-term forecasting.
Computational savings in training time and memory make embedding-free variants attractive for memory-constrained or resource-limited deployment.
The result calls for empirical justification of any new embedding component introduced in future forecasting architectures.

Reading between the lines

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

A natural extension is to test whether the same conclusion holds for other architectural components, such as normalization layers (for example, RevIN) or residual connections, since the paper explicitly leaves those interactions out of scope.
The finding suggests that learned embeddings may act as a form of overparameterization or implicit regularization rather than as necessary feature extractors; a controlled study of variance and training dynamics would clarify the mechanism.
Because the evaluation is limited to four regularly sampled ETT datasets, the result's scope is uncertain for irregularly sampled series, high-dimensional exogenous inputs, or non-stationary regimes.
A direct test would compare the permutation-and-concatenation replacement against a randomly initialized fixed linear projection of matched output dimension; if the improvements vanish, the reshaping itself, not the absence of learned embeddings, would be the active ingredient.

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.

Desk Editor's Note

read the letter

Colleague,

The headline: this is the first systematic ablation of embedding layers across a broad set of modern time series models, and the directional finding is probably right — embeddings often don't help and sometimes hurt. But the paper's central causal claim is undercut by a design flaw: the "without embedding" condition is not a clean deletion. It replaces the embedding with ad hoc permutation/concatenation preprocessing to satisfy the model's expected input shape. That makes the comparison a different model, not just the original minus an embedding. The stress-test note is on target here, and the paper never specifies these operations per model or validates that they are information-preserving. So we can't cleanly attribute the gains to the absence of embeddings.

What's genuinely good: the breadth — fifteen models, four standard datasets, reproduction of official code, and a consistent trend across most configurations. The authors also include an explicit limitations section and are careful to scope their claim to forecasting on regularly sampled data. The computational efficiency story (runtime and memory) is a useful practical addition, even if it's partly confounded by the same preprocessing change.

The soft spots beyond the main confound: most tables report single runs; confidence intervals are computed for only four models on two datasets, and Table 6 contains an impossible interval (MSE 0.392 with CI (0.394, 0.455)). That kind of internal inconsistency makes the statistical claims hard to trust. Also, all four datasets are ETT variants — same domain, same sensors, regular sampling. The "gains exceed SOTA differences" point is evocative but rests on a small convenience table.

Net assessment: the paper addresses an important question and the direction of the result is plausible, but it needs a clean ablation protocol and proper uncertainty quantification before the quantitative claims can be accepted. Worth sending to referees, with the expectation of major revision.

Recommendation: send to peer review — a serious referee could push the authors to either specify and verify the replacement preprocessing or redesign the ablation to isolate embeddings (e.g., keep the same projection but make it identity, random, or untrained).

Referee Report

3 major / 5 minor

Summary. This paper investigates whether data embedding layers in modern time series forecasting models are necessary. The authors take fifteen published forecasting models spanning Transformer-, MLP-, and hybrid/decomposition-based architectures, run them with their original embedding layers on four ETT benchmark datasets at horizons 96, 192, 336, and 720, and compare against variants with the embedding layers "bypassed" using permutation and concatenation preprocessing. The central claim is that removing data embedding layers usually does not degrade forecasting accuracy and in many cases improves both accuracy and computational efficiency; the paper further claims that these gains can exceed the performance differences typically reported between competing state-of-the-art models. The main evidence is presented as MSE/MAE tables (Tables 2-5), runtime and memory measurements, and confidence intervals for a subset of models in Table 6.

Significance. If the central claim survives scrutiny, it is a practically valuable negative result: it challenges a commonly used architectural component and could motivate simpler and cheaper forecasting pipelines. The breadth of the study—fifteen models, four datasets, two accuracy metrics, and runtime/memory measurements—is a genuine strength, as is the effort to provide code and to include confidence intervals for selected models. However, the current manuscript does not yet establish the causal claim that removing embeddings produces the observed improvements, because the no-embedding intervention is under-specified and the confidence-interval evidence contains internal contradictions. The paper's value is therefore conditional on a careful revision of the ablation protocol and the statistical reporting.

major comments (3)

[Section 5, "Configurations with degraded performance"; Section 1] The no-embedding condition is not a clean deletion of the embedding layer. The manuscript states that "in the absence of embedding layers, some preprocessing steps, such as permutation and concatenation are performed to reconcile the input with the model expected dimensions," and later that the raw input is "manually permutes, concatenates, and processes." Because the exact operations are never specified per model, the differences in Tables 2-5 cannot be attributed to the embedding layer alone. For models whose inverted embedding projects a length-96 sequence to a dimension-512 latent space, concatenation or tiling to fill the expected input shape effectively feeds repeated copies of the input to downstream layers, which is not equivalent to simply removing the embedding. The authors must specify the preprocessing for each of the fifteen models, demonstrate that it is information-preserving or otherwise controlled for input shape, and ideally validate the replacement against a simple fixed baseline such as a linear projection with a near-identity initialization.
[Section 5, "Confidence intervals"; Table 6] The claim that the confidence intervals for with-embedding and without-embedding models "do not overlap" is contradicted by the numbers in Table 6. For ETTh1, Times2D intervals (0.436, 0.442) and (0.429, 0.439) overlap; PDF intervals (0.451, 0.468) and (0.445, 0.459) overlap; and SOFTS intervals also overlap. Additionally, Table 6 contains internally impossible intervals: PDF on ETTm1 with embedding reports MSE 0.392 with 95% CI (0.394, 0.455), and SOFTS on ETTm1 with embedding reports MSE 0.408 with CI (0.403, 0.394). The statistical-significance argument in the text is therefore invalid and must be recomputed and restated; the main tables alone do not currently provide evidence that the improvements are statistically reliable.
[Section 4 and Tables 2-5] The manuscript does not specify how many independent runs are used for the main results in Tables 2-5. Several reported improvements are as small as 0.001-0.003 in MSE or MAE, and some configurations show degradation; without a seed count or variance information, single-run numbers cannot support the aggregate claim that removing embeddings improves accuracy in "over 95% of" configurations. The paper should state the number of runs and report error bars or standard deviations for all principal comparisons, or explicitly restrict significance claims to the confidence-interval experiments.

minor comments (5)

[Section 5, "Accuracy typically improves without embeddings"] The claim that improvements occur in "over 95%" of configurations is not backed by a precise count or a definition of a configuration. Some table entries show degradation (e.g., Crossformer on ETTh1 at H=96, MICN on ETTm1 at H=96, PatchTST on ETTm1 at H=96); the authors should provide the exact numerator and denominator and state whether a configuration means a model-horizon pair or a model-horizon-metric cell.
[Table 6 and Appendix A.4] There are numerous formatting and transcription errors in the confidence-interval tables: the header "Time2D" should read "Times2D," Table 6 contains impossible intervals as noted above, and Tables 8-9 contain malformed entries such as "0.461 0.458, 0.464)" and extra digits like "0.4152" and "0.5532." These errors undermine reader trust in the numerical results and should be corrected throughout.
[Section 1 and Appendix A.1] The limitations section does not mention that the no-embedding condition relies on permutation/concatenation preprocessing, which is a potential confound for the paper's main causal claim. This limitation should be acknowledged explicitly and, ideally, addressed with sensitivity analyses.
[Section 4 and Table 1] Table 1 lists iTransformer, RLinear, and TimeMixer, but these models do not appear in the ablation tables; the paper should clarify whether they are excluded from the fifteen evaluated models or are only reference points, and, if excluded, state the reason.
[Abstract and GitHub link] The GitHub repository name "neuripsdataembedidng" appears to contain a typo and should be verified; the link should point to a stable, accessible repository since the manuscript relies on it for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical ablation whose central claim does not reduce to its inputs or to a self-citation chain.

full rationale

The paper presents an empirical ablation of data embedding layers across fifteen forecasting models; it contains no derivation in which a predicted quantity is equivalent by construction to a fitted input. The 'without embedding' condition is implemented by bypassing embedding layers and, where needed, permuting/concatenating raw inputs to match expected dimensions (Section 1 and Section 5). This preprocessing is a potential confound for the causal attribution to embedding removal, and the limitations section (A.1) does not explicitly list it; however, that is a construct-validity or reproducibility concern, not a circularity, because the reported accuracy and efficiency numbers are not derived from the preprocessing rule or from the paper's own definitions. The only self-citation is Times2D [15] as one of fifteen baselines; the central finding does not depend on that model alone, and Times2D is externally evaluated against the other published models in Table 1. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from prior work, and no ansatz is smuggled in via self-citation. The central claim would stand or fall on the experimental comparison regardless of how the no-embedding variant is constructed, so the derivation chain is self-contained.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No free parameters are fitted in this study; the models use their original hyperparameters and the ETT datasets are fixed benchmarks. The central claim rests on the faithfulness of the embedding-removal procedure, the representativeness of the four ETT datasets, and the assumption that official baseline reproductions are accurate. No new entities are introduced.

assumptions (3)

ad hoc to paper Bypassing embedding layers and replacing them with permutation/concatenation of raw inputs isolates the effect of the embedding layer.
Invoked in Section 1 and Section 5; no validation is provided that this replacement is the only meaningful change.
domain assumption The four ETT datasets are representative of multivariate time series forecasting tasks.
Section 4 uses only ETTh1, ETTh2, ETTm1, and ETTm2; the conclusion is framed generally for time series forecasting.
domain assumption Baseline reproductions using official repositories match the original papers' settings.
Section 4 states reproduction using official code; no comparison to published numbers is shown for all fifteen models.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Are Data Embeddings effective in time series forecasting?." pith.science (2026). https://pith.science/paper/XRA7MSPJ

@misc{pith2026250520716,
  author       = {Pith},
  title        = {Pith review of: Are Data Embeddings effective in time series forecasting?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XRA7MSPJ}},
  note         = {Machine review of arXiv:2505.20716}
}

read the original abstract

Time series forecasting plays a crucial role in many real-world applications, and numerous complex forecasting models have been proposed in recent years. Despite their architectural innovations, most state-of-the-art models report only marginal improvements -- typically just a few thousandths in standard error metrics. These models often incorporate complex data embedding layers to transform raw inputs into higher-dimensional representations to enhance accuracy. But are data embedding techniques actually effective in time series forecasting? Through extensive ablation studies across fifteen state-of-the-art models and four benchmark datasets, we find that removing data embedding layers from many state-of-the-art models does not degrade forecasting performance. In many cases, it improves both accuracy and computational efficiency. The gains from removing embedding layers often exceed the performance differences typically reported between competing models. Code available at: https://github.com/neuripsdataembedidng/DataEmbedding

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 37 canonical work pages

[1]

Edformer: Embedded decom- position transformer for interpretable multivariate time series predictions, 2024

Sanjay Chakraborty, Ibrahim Delibasoglu, and Fredrik H eintz. Edformer: Embedded decom- position transformer for interpretable multivariate time series predictions, 2024

work page 2024
[2]

Con- tiformer: Continuous-time transformer for irregular time series modeling

Y uqi Chen, Kan Ren, Y ansen Wang, Y uchen Fang, Weiwei Sun, and Dongsheng Li. Con- tiformer: Continuous-time transformer for irregular time series modeling. In Thirty-seventh Conference on Neural Information Processing Systems , 2023

work page 2023
[3]

Pe- riodicity decoupling framework for long-term series forec asting

Tao Dai, Beiliang Wu, Peiyuan Liu, Naiqi Li, Jigang Bao, Y ong Jiang, and Shu-Tao Xia. Pe- riodicity decoupling framework for long-term series forec asting. In The Twelfth International Conference on Learning Representations, 2024

work page 2024
[4]

Pe- riodicity decoupling framework for long-term series forec asting

Tao Dai, Beiliang Wu, Peiyuan Liu, Naiqi Li, Jigang Bao, Y ong Jiang, and Shu-Tao Xia. Pe- riodicity decoupling framework for long-term series forec asting. International Conference on Learning Representations, 2024

work page 2024
[5]

SOFT S: Efﬁcient multivariate time series forecasting with series-core fusion

Lu Han, Xu-Y ang Chen, Han-Jia Y e, and De-Chuan Zhan. SOFT S: Efﬁcient multivariate time series forecasting with series-core fusion. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

work page 2024
[6]

WITRAN: Water-wave information transmission and recurrent accele ration network for long-range time series forecasting

Y uxin Jia, Y oufang Lin, Xinyan Hao, Y an Lin, Shengnan Guo , and Huaiyu Wan. WITRAN: Water-wave information transmission and recurrent accele ration network for long-range time series forecasting. In Thirty-seventh Conference on Neural Information Processi ng Systems , 2023

work page 2023
[7]

Zha ng, Xiaoming Shi, Pin-Y u Chen, Y uxuan Liang, Y uan-Fang Li, Shirui Pan, and Qingsong Wen

Ming Jin, Shiyu Wang, Lintao Ma, Zhixuan Chu, James Y . Zha ng, Xiaoming Shi, Pin-Y u Chen, Y uxuan Liang, Y uan-Fang Li, Shirui Pan, and Qingsong Wen. Time-LLM: Time series forecasting by reprogramming large language models. In The Twelfth International Conference on Learning Representations, 2024

work page 2024
[8]

V ardrop: Enh ancing training efﬁciency by reducing variate redundancy in periodic time series foreca sting

Junhyeok Kang, Y ooju Shin, and Jae-Gil Lee. V ardrop: Enh ancing training efﬁciency by reducing variate redundancy in periodic time series foreca sting. In AAAI, 2025

work page 2025

Show all 39 references

[9]

Towards localization via data embedding for tabPFN

Mykhailo Koshil, Thomas Nagler, Matthias Feurer, and Ka tharina Eggensperger. Towards localization via data embedding for tabPFN. In NeurIPS 2024 Third T able Representation Learning W orkshop, 2024

2024
[10]

Lagcnn: A fast yet effective model f or multivariate long-term time series forecasting

Linsen Li, Chunfei Jian, Feng Wan, Dongdong Geng, Ziqua n Fang, Lu Chen, Y unjun Gao, Wei- hao Jiang, and Jiang Zhu. Lagcnn: A fast yet effective model f or multivariate long-term time series forecasting. In Proceedings of the 33rd ACM International Conference on Inf ormation ...

2024
[11]

Revisiting lo ng-term time series forecasting: An investigation on afﬁne mapping, 2024

Zhe Li, Shiyi Qi, Yiduo Li, and Zenglin Xu. Revisiting lo ng-term time series forecasting: An investigation on afﬁne mapping, 2024

2024
[12]

Multivariate time series anomaly detection and interpretation using hie rarchical inter-metric and temporal embedding

Zhihan Li, Y oujian Zhao, Jiaqi Han, Y a Su, Rui Jiao, Xida o Wen, and Dan Pei. Multivariate time series anomaly detection and interpretation using hie rarchical inter-metric and temporal embedding. In Proceedings of the 27th ACM SIGKDD conference on knowledge d iscovery & dat...

2021
[13]

Minus- former: Improving time series forecasting by progressivel y learning residuals

Daojun Liang, Haixia Zhang, Dongfeng Y uan, Bingzheng Z hang, and Minggao Zhang. Minus- former: Improving time series forecasting by progressivel y learning residuals. arXiv preprint arXiv:2402.02332, 2024

2024 arXiv
[14]

itransformer: Inverted transformers are effective f or time series forecasting

Y ong Liu, Tengge Hu, Haoran Zhang, Haixu Wu, Shiyu Wang, Lintao Ma, and Mingsheng Long. itransformer: Inverted transformers are effective f or time series forecasting. arXiv preprint arXiv:2310.06625, 2023

2023 arXiv
[15]

Times2d: Multi-period de- composition and derivative mapping for general time series forecasting

Reza Nematirad, Anil Pahwa, and Balasubramaniam Natar ajan. Times2d: Multi-period de- composition and derivative mapping for general time series forecasting. In Proceedings of the AAAI Conference on Artiﬁcial Intelligence , volume 39, pages 19651–19658, 2025. 11

2025
[16]

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

Y uqi Nie, Nam H Nguyen, Phanwadee Sinthong, and Jayant K alagnanam. A time series is worth 64 words: Long-term forecasting with transformers. I n The Eleventh International Con- ference on Learning Representations , 2023

2023
[17]

Ppdformer: Channel-speci ﬁc periodic patch division for time series forecasting

Meng Wan, Qi Su, Huan Hao, Jue Wang, Y uexiu Cui, Y uxuan Bi , Rongqiang Cao, Peng Shi, Y angang Wang, Zonghua Qiu, et al. Ppdformer: Channel-speci ﬁc periodic patch division for time series forecasting. In ICASSP 2025-2025 IEEE International Conference on Acousti cs, Speech a...

2025
[18]

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

Huiqiang Wang, Jian Peng, Feihu Huang, Jince Wang, Junh ui Chen, and Yifei Xiao. MICN: Multi-scale local and global context modeling for long-term series forecasting. In The Eleventh International Conference on Learning Representations , 2023

2023
[19]

Zhang, and JUN ZHOU

Shiyu Wang, Haixu Wu, Xiaoming Shi, Tengge Hu, Huakun Lu o, Lintao Ma, James Y . Zhang, and JUN ZHOU. Timemixer: Decomposable multiscale mixing fo r time series forecasting. In The Twelfth International Conference on Learning Represen tations, 2024

2024
[20]

Timexer: Empowering transformers for time series forecasting with exogenous variables

Y uxuan Wang, Haixu Wu, Jiaxiang Dong, Guo Qin, Haoran Zh ang, Y ong Liu, Y un-Zhong Qiu, Jianmin Wang, and Mingsheng Long. Timexer: Empowering transformers for time series forecasting with exogenous variables. In The Thirty-eighth Annual Conference on Neural Information Proce...

2024
[21]

ETSformer: Ex- ponential smoothing transformers for time-series forecas ting, 2023

Gerald Woo, Chenghao Liu, Doyen Sahoo, Akshat Kumar, an d Steven Hoi. ETSformer: Ex- ponential smoothing transformers for time-series forecas ting, 2023

2023
[22]

Lino: Advancing recursive residual decomposition of linea r and nonlinear patterns for robust time series forecasting, 2025

Guoqi Y u, Y aoming Li, Xiaoyu Guo, Shujun Wang, Zirui Liu , Dayu Wang, and Tong Y ang. Lino: Advancing recursive residual decomposition of linea r and nonlinear patterns for robust time series forecasting, 2025

2025
[23]

Crossformer: Transformer utilizing cross-dimension depen- dency for multivariate time series forecasting

Y unhao Zhang and Junchi Y an. Crossformer: Transformer utilizing cross-dimension depen- dency for multivariate time series forecasting. In The Eleventh International Conference on Learning Representations, 2023. 12 A Technical Appendices and Supplementary Material A.1 Limitat...

2023
[24]

Claims Question: Do the main claims made in the abstract and introdu ction accurately reﬂect the paper’s contributions and scope? Answer: [Y es] Justiﬁcation: The abstract and introduction clearly state the paper’s contributions and scope
[25]

Limitations Question: Does the paper discuss the limitations of the work performed by the au- thors? Answer: [Y es] Justiﬁcation: The limitations of the study are explicitly d iscussed in Section A.1
[26]

Theory assumptions and proofs Question: For each theoretical result, does the paper provi de the full set of assumptions and a complete (and correct) proof? Answer: [NA] Justiﬁcation: This paper does not include theoretical resu lts or formal proofs, as it is fo- cused on empi...
[27]

Experimental result reproducibility Question: Does the paper fully disclose all the information needed to reproduce the main experimental results of the paper to the extent that it affec ts the main claims and/or conclu- sions of the paper (regardless of whether the code and d...
[28]

Open access to data and code Question: Does the paper provide open access to the data and c ode, with sufﬁcient instruc- tions to faithfully reproduce the main experimental result s, as described in supplemental material? Answer: [Y es] Justiﬁcation: The code is provided as an...
[29]

The full implementation is provided in the sup plementary material and the anonymous GitHub repository

Experimental setting/details Question: Does the paper specify all the training and test de tails (e.g., data splits, hyper- parameters, how they were chosen, type of optimizer, etc.) n ecessary to understand the results? Answer: [Y es] Justiﬁcation: Key training and evaluation...
[30]

Experiment statistical signiﬁcance Question: Does the paper report error bars suitably and corr ectly deﬁned or other appropri- ate information about the statistical signiﬁcance of the ex periments? Answer: [Y es] Justiﬁcation: The conﬁdence intervals are provided in Sect ions...
[31]

Experiments compute resources 16 Question: For each experiment, does the paper provide sufﬁc ient information on the com- puter resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [Y es] Justiﬁcation: The compute resource...
[32]

Code of ethics Question: Does the research conducted in the paper conform, in every respect, with the NeurIPS Code of Ethics https://neurips.cc/public/EthicsGuidelines? Answer: [Y es] Justiﬁcation: The research complies with the NeurIPS Code o f Ethics
[33]

Broader impacts Question: Does the paper discuss both potential positive so cietal impacts and negative societal impacts of the work performed? Answer: [Y es] Justiﬁcation: We have discussed the broader and societal im pacts in Section A.2
[34]

Safeguards Question: Does the paper describe safeguards that have been put in place for responsible release of data or models that have a high risk for misuse (e.g ., pretrained language models, image generators, or scraped datasets)? Answer: [NA] Justiﬁcation: The paper does ...
[35]

Licenses for existing assets Question: Are the creators or original owners of assets (e.g ., code, data, models), used in the paper, properly credited and are the license and terms of use explicitly mentioned and properly respected? Answer: [Y es] Justiﬁcation: We will release...
[36]

New assets Question: Are new assets introduced in the paper well docume nted and is the documenta- tion provided alongside the assets? Answer: [Y es] Justiﬁcation: The assets are included in the anonymized rep ository and the attached supple- mentary zip ﬁle
[37]

Crowdsourcing and research with human subjects Question: For crowdsourcing experiments and research with human subjects, does the pa- per include the full text of instructions given to participa nts and screenshots, if applicable, as well as details about compensation (if any)...
[38]

Institutional review board (IRB) approvals or equivalent f or research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Insti tutional Review Board (IRB) approva...
[39]

Declaration of LLM usage 17 Question: Does the paper describe the usage of LLMs if it is an important, original, or non-standard component of the core methods in this research ? Answer: [NA] Justiﬁcation: LLMs were not used as an original or non-stand ard component in the core...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Edformer: Embedded decom- position transformer for interpretable multivariate time series predictions, 2024

Sanjay Chakraborty, Ibrahim Delibasoglu, and Fredrik H eintz. Edformer: Embedded decom- position transformer for interpretable multivariate time series predictions, 2024

work page 2024

[2] [2]

Con- tiformer: Continuous-time transformer for irregular time series modeling

Y uqi Chen, Kan Ren, Y ansen Wang, Y uchen Fang, Weiwei Sun, and Dongsheng Li. Con- tiformer: Continuous-time transformer for irregular time series modeling. In Thirty-seventh Conference on Neural Information Processing Systems , 2023

work page 2023

[3] [3]

Pe- riodicity decoupling framework for long-term series forec asting

Tao Dai, Beiliang Wu, Peiyuan Liu, Naiqi Li, Jigang Bao, Y ong Jiang, and Shu-Tao Xia. Pe- riodicity decoupling framework for long-term series forec asting. In The Twelfth International Conference on Learning Representations, 2024

work page 2024

[4] [4]

Pe- riodicity decoupling framework for long-term series forec asting

Tao Dai, Beiliang Wu, Peiyuan Liu, Naiqi Li, Jigang Bao, Y ong Jiang, and Shu-Tao Xia. Pe- riodicity decoupling framework for long-term series forec asting. International Conference on Learning Representations, 2024

work page 2024

[5] [5]

SOFT S: Efﬁcient multivariate time series forecasting with series-core fusion

Lu Han, Xu-Y ang Chen, Han-Jia Y e, and De-Chuan Zhan. SOFT S: Efﬁcient multivariate time series forecasting with series-core fusion. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

work page 2024

[6] [6]

WITRAN: Water-wave information transmission and recurrent accele ration network for long-range time series forecasting

Y uxin Jia, Y oufang Lin, Xinyan Hao, Y an Lin, Shengnan Guo , and Huaiyu Wan. WITRAN: Water-wave information transmission and recurrent accele ration network for long-range time series forecasting. In Thirty-seventh Conference on Neural Information Processi ng Systems , 2023

work page 2023

[7] [7]

Zha ng, Xiaoming Shi, Pin-Y u Chen, Y uxuan Liang, Y uan-Fang Li, Shirui Pan, and Qingsong Wen

Ming Jin, Shiyu Wang, Lintao Ma, Zhixuan Chu, James Y . Zha ng, Xiaoming Shi, Pin-Y u Chen, Y uxuan Liang, Y uan-Fang Li, Shirui Pan, and Qingsong Wen. Time-LLM: Time series forecasting by reprogramming large language models. In The Twelfth International Conference on Learning Representations, 2024

work page 2024

[8] [8]

V ardrop: Enh ancing training efﬁciency by reducing variate redundancy in periodic time series foreca sting

Junhyeok Kang, Y ooju Shin, and Jae-Gil Lee. V ardrop: Enh ancing training efﬁciency by reducing variate redundancy in periodic time series foreca sting. In AAAI, 2025

work page 2025