Pith. sign in

REVIEW 3 major objections 3 minor 42 references

REATS: LLM Reasoning-based Ensemble Learning for Adaptive Time Series Forecasting

T0 review · 3 major / 3 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read A 1.7B-parameter language model, fine-tuned with rule-based reasoning and a reciprocal reward, assigns per-sample ensemble weights that beat fixed, neural, and zero-shot LLM baselines on all eight forecasting benchmarks.

desk verdict A solid, well-engineered LLM ensemble router for time series with a thoughtful reward design; the main unresolved question is how much of the gain comes from the LLM rather than the retrieval prior. read the letter →

arxiv 2608.10149 v1 pith:ZQCQZUFZ submitted 2026-08-10 cs.LG

Xu Zhang , Chang Xu , Hui Sun , Nan Ma , Zijian Zhang , Peng Wang , Wei Wang , Li Zhao This is my paper · ORCID
classification cs.LG
keywords timeseriesforecastingensemblelearninglargelanguagemodelschain-of-thoughtreasoningGRPOrewarddesignsample-adaptiveweightingout-of-domaingeneralization
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 claims that a 1.7-billion-parameter language model, fine-tuned in two stages, can act as a sample-adaptive ensemble router for time series forecasting: given a hybrid textual–numerical description of a series and a set of candidate forecasters, it produces per-sample ensemble weights that outperform fixed weighting, neural-network routers, and zero-shot LLM prompting on all eight benchmarks tested. The authors argue that the key enablers are a fixed-token representation of the series, rule-generated chain-of-thought reasoning that needs no API calls, diverse multi-row weight supervision, and a reciprocal reward mapping that adapts GRPO to a continuous unbounded regression objective. If the claim holds, LLM reasoning becomes a practical and interpretable tool for model combination, including transfer to unseen candidate models.

What carries the argument

The machinery is the combination of a structured prompt and a reciprocal reward mapping. The structured prompt renders the series as fixed-token hybrid text (eight feature groups as key-value pairs) together with retrieved similar-sample priors, which makes rule-based chain-of-thought construction possible without API calls. The reciprocal reward mapping $r = 1/(1+k\delta)$ with $k=20$ compresses the unbounded MSE gap $\delta$ into $[0,1]$ with polynomial decay, preserving near-oracle sensitivity after GRPO's advantage normalization, unlike the naive $r=-\delta$ whose outlier-dominated variance collapses the normalized advantages of near-optimal candidates. A ten-row integer-percentage weight table (one oracle row plus nine diverse Dirichlet-sampled rows) provides multi-row supervision and denser rewards during both SFT and GRPO.

What would settle it

Train REATS-GRPO with oracle weights replaced by validation-set weights or by random but diverse weights on the same benchmarks; if the reported margin over the best baselines does not collapse, the advantage does not depend on the oracle-quality assumption. A second decisive probe is to fit a non-reasoning regressor (for example gradient-boosted trees) on the same eight feature groups and RAG statistics; if it matches or beats REATS-GRPO, the active ingredient is the features, not LLM reasoning.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that an LLM router, REATS, assigns ensemble weights by reasoning over eight feature groups of the time series (stationarity, noise, trend, seasonality, autocorrelation, statistics, outliers, distribution) plus retrieved similar training samples, and this beats the strongest traditional baselines: average MSE of 0.1384 on foundation-model candidates versus 0.1597 for optimal validation-set weighting, and 0.1080 versus 0.1352 on small-model candidates, winning on all eight datasets in each group. The same router, trained only on small candidates, generalizes to unseen foundation models (0.1442 vs. 0.1564 best baseline). The paper attributes the gain to two-stage fine-tuning in which SFT imitates rule-based reasoning and GRPO then directly optimizes a reciprocal-transformed MSE reward, with the diverse ten-row weight table supplying denser reward signals.

Load-bearing premise

The load-bearing premise is that per-sample oracle weights computed from training-set ground truth are stable and informative enough that a model can learn a mapping from the eight feature groups to those weights and have it transfer to test samples; if the oracle quadratic program overfits training noise, both the SFT labels and the GRPO rewards inherit that bias.

Editorial extensions

If this is right

  • Sample-adaptive ensembling can be driven by explicit reasoning rather than fixed rules, and the weights are explainable in natural language.
  • Rule-based chain-of-thought, generated without paid API calls, is sufficient for SFT and is critical for out-of-domain generalization to unseen candidate models.
  • A small 1.7B fine-tuned LLM outperforms zero-shot much larger LLMs on this routing task, so scale is not the decisive factor.
  • The reciprocal reward mapping makes GRPO applicable to continuous regression objectives, outperforming optimization-level GRPO variants (DAPO, DrGRPO, GSPO, SAPO) that keep the naive unbounded reward.
  • The router transfers across datasets and to unseen candidate model sets, suggesting it learns semantic temporal reasoning rather than memorizing model identities.

Reading between the lines

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

  • The authors do not test, but the same router design should apply to other continuous model-selection problems (e.g., per-sample classifier blending or hyperparameter recommendation) whenever an offline oracle can be computed from training data.
  • Because the input token cost is fixed regardless of series length, the approach should scale to very long histories without truncation; this is a direct consequence the authors note but do not test at length 512 in the main results.
  • A decisive comparison the authors do not run: a non-LLM regressor on the same eight feature groups would isolate whether LLM reasoning or feature engineering drives the gain.
  • The reciprocal mapping's benefit is predicted to grow as GRPO rollout groups become more mixed in quality; measuring within-group reward variance under $r=-\delta$ versus $r=1/(1+k\delta)$ would test the authors' mechanism directly.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. REATS proposes to replace fixed or black-box ensemble rules for time series forecasting with a fine-tuned 1.7B LLM that outputs sample-adaptive ensemble weights from hybrid textual-numerical features, retrieved similar-case priors, and chain-of-thought reasoning. The pipeline constructs rule-based CoT supervision from per-sample oracle weights computed by a QP on training ground truth, formats weights as integer percentage tables, and trains with SFT followed by GRPO using a reciprocal reward mapping that compresses MSE gaps. Experiments on eight univariate forecasting benchmarks with four foundation-model or four small-model candidates report that REATS-GRPO achieves the lowest average MSE in both model groups and in OOD transfer to unseen candidates, while ablations attribute gains to CoT, RAG, table format, and multi-row supervision.

Significance. The framework is timely and the OOD experiments are carefully constructed: candidate descriptions and the RAG pool are rebuilt from training-set predictions and ground truth, so no test-label leakage is apparent. Rule-CoT removes API dependency with comparable quality, and the reciprocal reward analysis in Appendix A.5.3 provides a concrete, testable account of why naive r=-delta compresses GRPO advantages. If the attribution gap described below can be closed, REATS would be a meaningful step toward interpretable LLM-based ensemble routers that adapt to candidate sets without retraining.

major comments (3)
  1. [Section 3.3, Algorithm 2, Appendix A.8] The central claim that fine-tuned LLM reasoning drives the MSE gains is not yet established, because the inference prompt already contains oracle-derived weights for the retrieved training samples. The prompt in Appendix A.8 includes 'Optimal ensemble weights(%) on the most similar time series:', and Algorithm 2 Step 3 instructs the CoT to state agreement or divergence from the retrieved average; the model is SFT-trained to imitate oracle rows, so it may simply adjust the retrieved average. The 'w/o RAG' ablation in Table 3(b) removes both retrieval and the weight prior, so it does not control for the prior alone. Please add baselines that (i) output the retrieved average weight directly (retrieval-only) and (ii) regress the eight feature groups onto oracle weights with a ridge or kNN predictor, evaluated in the same ID and OOD settings, and report the margin of REATS over these controls.
  2. [Tables 2-6 and Figure 6] No repeated-seed statistics are reported, and the design parameters K, K', k, lambda1/lambda3, and the format penalty appear to be selected on the same eight benchmarks used for the headline comparisons (e.g., Table 4 picks K'=10, Figure 6 picks the default reward coefficients, and Appendix A.5.1 picks k=20). Thus the current evidence cannot separate genuine improvement from selection on the reported test sets. Please provide mean and standard deviation over at least three seeds for REATS and the strongest baselines, and move hyperparameter selection to a validation-based protocol.
  3. [Section 4.3, Figure 5(d), Table 8] The comparison with DAPO, DrGRPO, GSPO, and SAPO covers only three datasets (Exchange, Electricity, Weather), while the claim that the reciprocal reward mapping outperforms several recent GRPO algorithmic variants is stated generally. Moreover, the variants are only tested with the naive r=-delta reward, so the interaction between optimizer-level and reward-level fixes is not measured. Please extend the comparison to the full eight benchmarks and include a factorial design (naive vs reciprocal reward, with standard GRPO and at least one variant) to support the stated claim.
minor comments (3)
  1. [Section 2 and Table 2(a)] There is a typo in Section 2: 'The first istraditional machine learning-based methods' should read 'The first is traditional machine learning-based methods'; Table 2(a) also prints 'NNRLM C' instead of 'RLMC'.
  2. [Section 4.1] The implementation details do not report the GRPO group size G, sampling temperature, learning rate, number of GRPO steps, or training compute; adding these would substantially improve reproducibility.
  3. [Appendix A.8] The RAG reference block in the SFT example shows three rows of retrieved weights while the method states K'=10 rows; please clarify whether the display is truncated or whether the retrieval pool stores fewer rows than the supervision table.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found; the oracle-weighted supervision is a legitimate supervised target and no test-time quantity is defined in terms of the quantity being predicted.

full rationale

The derivation chain is self-contained. Per-sample oracle weights w* (Eq. 3 and the subsequent QP) are computed from training-set ground truth and candidate prediction errors, then used as SFT labels and GRPO reward references; this is ordinary supervised learning, not a circular prediction. At test time, REATS generates weights from hybrid features and retrieved training samples without access to test labels, and the paper states this explicitly. The SFT example (Appendix A.8) shows the model output row is not identical to the retrieved oracle row, so the output is not forced by construction. The OptW baseline transparently uses the same QP objective as a fixed global weight, which is a fair comparison rather than a hidden reduction. Citations to the authors' own prior models (LSINet, SEMixer, MLF) are ordinary candidate pool choices and are not load-bearing premises, uniqueness theorems, or ansatz justifications. The only substantive concern is that the RAG prior injects oracle-derived weights into the prompt and no retrieval-only baseline is evaluated, which bears on whether the LLM's reasoning, rather than the retrieval prior, drives the gains; that is an experimental-isolation issue, not circularity under the strict standard of exhibiting an equation or fitted parameter that is equivalent to the claimed prediction by definition.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

No new physical entities are introduced. The central method is an empirical supervised pipeline: it uses standard linear algebra for oracle weights, a domain assumption that training-set oracle weights transfer, a design assumption that eight feature groups capture routing-relevant structure, and several tuned hyperparameters (k, K, K', reward coefficients). The rule-based CoT is generated from the same oracle weights used as labels, which is a self-consistency choice but not a circular derivation.

free parameters (5)
  • Reward scale k = 20 (valid range [10,25])
    Scaling factor in reciprocal reward r = 1/(1 + k*delta); chosen via heuristic criteria on near-oracle sensitivity and mid-range signal, not from a principled derivation.
  • Retrieval neighbors K = 3
    Selected by comparing k=3 vs k=25 on the small-model group; k=3 gives lower average MSE and is set as default.
  • Supervision rows K' = 10
    Nine diverse Dirichlet-sampled rows plus the oracle row; ablation shows K'=10 beats K'=1 in GRPO.
  • Reward coefficients (lambda1, lambda2, lambda3) = (0.8, 0, 0.2)
    Selected among six configurations in Figure 6 on the same benchmarks; G1 is the default.
  • Format penalty = -0.5
    Hand-set negative penalty for format violations such as negative weights or sums not equal to 100%.
assumptions (4)
  • standard math The closed-form oracle weight solution requires invertibility of G, handled by regularizing G with epsilon*I.
    Appendix A.6, Eq. (3): the unconstrained solution uses G^{-1}; the paper adds epsilon=1e-10 to avoid ill-conditioning.
  • domain assumption Per-sample oracle weights computed from training-set ground truth are a valid supervision signal for test-time routing.
    Section 3.3 states all oracle weights for SFT and GRPO are derived from training set ground truth only; if these weights do not transfer, the whole pipeline is biased.
  • domain assumption The eight hand-crafted feature groups capture the temporal characteristics relevant to ensemble weighting.
    Section 3.3, Hybrid textual-numerical feature extraction and Appendix Table 13: stationarity, noise, trend, seasonality, autocorrelation, statistics, outliers, and distribution are assumed sufficient for routing decisions.
  • ad hoc to paper Rule-based CoT templates that reverse-engineer oracle weights produce supervision of comparable quality to LLM-generated CoT.
    Algorithm 1 and Appendix Algorithm 2 generate reasoning from the same oracle weights; the paper validates this empirically in Table 6(a), but the templates are not derived from an external theory.

how reviews work

0 comments
Cite this review

Pith. "Pith review of REATS: LLM Reasoning-based Ensemble Learning for Adaptive Time Series Forecasting." pith.science (2026). https://pith.science/paper/ZQCQZUFZ

@misc{pith2026260810149,
  author       = {Pith},
  title        = {Pith review of: REATS: LLM Reasoning-based Ensemble Learning for Adaptive Time Series Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZQCQZUFZ}},
  note         = {Machine review of arXiv:2608.10149}
}
read the original abstract

Due to the diversity of real-world time series, no single forecasting model consistently dominates across all samples. Ensemble learning addresses this by combining complementary model strengths, yet existing methods rely on fixed rules or black-box models based solely on numerical inputs, failing to leverage LLM reasoning for interpretable weighting decisions. We propose REATS, which leverages LLM reasoning capabilities as an intelligent ensemble router that jointly processes textual temporal pattern descriptions and numerical features to produce interpretable, sample-adaptive ensemble weights through chain-of-thought reasoning. To enable effective LLM-based ensembling, we study its key design choices and propose: (i) a structured input pipeline that transforms raw time series into hybrid textual--numerical representations with fixed token cost, enabling rule-based chain-of-thought construction without API dependency, augmented with retrieved similar-sample priors; (ii) a diverse multi-row weight supervision scheme coupled with a token-efficient percentage-table format that reduces numerical complexity and mitigates LLM hallucinations; and (iii) a two-stage fine-tuning framework combining SFT with GRPO, where a reciprocal reward mapping transforms the continuous unbounded MSE gap into bounded signals with amplified near-oracle sensitivity, addressing the uniform sensitivity and outlier-dominated advantage compression inherent in naive reward designs for regression-based GRPO. Experiments on eight benchmarks demonstrate that REATS outperforms competitive ensemble baselines while providing natural language explanations and demonstrating strong transfer learning and out-of-domain generalization to unseen candidate models.

Figures

Figures reproduced from arXiv: 2608.10149 by the authors.

Figure 1
Figure 1. Overall framework of REATS. Hybrid textual–numerical feature extraction. How to effectively represent time series for LLM-based reasoning is a key design choice. Directly tokenizing raw values as text causes token counts to grow linearly with series length (e.g., 1656→7480 from length 96 to 512) and increases hallucination risk, while encoding series through a separate MLP projector sacrifices interpretability and O… view at source ↗
Figure 2
Figure 2. Structured LLM fine-tuning data construction pipeline. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Scalability to different numbers of candidate models. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Reward mapping analysis. (a) Raw reward sensitivity [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: GRPO reward coefficient sensitivity across [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Transfer learning comparison (small model group, MSE [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Reward curves under varying k for (a) exponential, (b) linear, and (c) reciprocal mapping functions. Bold lines indicate the selected k values. (d) Comparison of all four mappings at their recommended k. The derivatives of the three reward functions are: Reciprocal [P…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 29 canonical work pages

  1. [1]

    A new framework for multivariate time series forecasting in energy management system.IEEE Transactions on Smart Grid, 14(4):2934–2947, 2022

    Niko Uremovi ´c, Marko Bizjak, Primož Suki ˇc, Gorazd Štumberger, Borut Žalik, and Niko Luka ˇc. A new framework for multivariate time series forecasting in energy management system.IEEE Transactions on Smart Grid, 14(4):2934–2947, 2022. 11 REATS: LLM Reasoning-based Ensemble Learning for Adaptive Time Series Forecasting

    work page 2022

  2. [2]

    Multi-period learning for financial time series forecasting

    Xu Zhang, Zhengang Huang, Yunzhi Wu, Xun Lu, Erpeng Qi, Yunkai Chen, Zhongya Xue, Qitong Wang, Peng Wang, and Wei Wang. Multi-period learning for financial time series forecasting. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 1, pages 2848–2859, 2025

    work page 2025

  3. [3]

    Forecasts in urban transportation planning: Uses, methods, and dilemmas.Climatic Change, 11(1):61–80, 1987

    Martin Wachs. Forecasts in urban transportation planning: Uses, methods, and dilemmas.Climatic Change, 11(1):61–80, 1987

    work page 1987

  4. [4]

    A combined model for multivariate time series forecasting based on mlp-feedforward attention-lstm.IEEE Access, 10:88644–88654, 2022

    Yuntong Liu, Chunna Zhao, and Yaqun Huang. A combined model for multivariate time series forecasting based on mlp-feedforward attention-lstm.IEEE Access, 10:88644–88654, 2022

    work page 2022

  5. [5]

    A lightweight sparse interaction network for time series forecasting

    Xu Zhang, Qitong Wang, Peng Wang, and Wei Wang. A lightweight sparse interaction network for time series forecasting. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 13304–13312, 2025

    work page 2025

  6. [6]

    Drcnn: decomposing residual convolutional neural networks for time series forecasting.Scientific Reports, 13(1):15901, 2023

    Yuzhen Zhu, Shaojie Luo, Di Huang, Weiyan Zheng, Fang Su, and Beiping Hou. Drcnn: decomposing residual convolutional neural networks for time series forecasting.Scientific Reports, 13(1):15901, 2023

    work page 2023

  7. [7]

    RRMSE Voting Regressor: A weighting function based improvement to ensemble regression

    Shikun Chen and Nguyen Manh Luc. Rrmse voting regressor: A weighting function based improvement to ensemble regression.arXiv preprint arXiv:2207.04837, 2022

    work page Pith review arXiv 2022

  8. [8]

    Ensemble learning of inverse probability weights for marginal structural modeling in large observational datasets.Statistics in medicine, 34(1):106–117, 2015

    Susan Gruber, Roger W Logan, Inmaculada Jarrín, Susana Monge, and Miguel A Hernán. Ensemble learning of inverse probability weights for marginal structural modeling in large observational datasets.Statistics in medicine, 34(1):106–117, 2015

    work page 2015

Show all 42 references
  1. [9]

    Reinforcement learning based dynamic model combination for time series forecasting

    Yuwei Fu, Di Wu, and Benoit Boulet. Reinforcement learning based dynamic model combination for time series forecasting. InProceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 6639–6647, 2022

    2022

  2. [10]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300, 2024

    2024 arXiv

  3. [11]

    Specfuse: Ensembling large language models via next-segment prediction.arXiv preprint arXiv:2412.07380, 2024

    Bo Lv, Nayu Liu, Chen Tang, Xin Liu, Yue Yu, and Ping Luo. Specfuse: Ensembling large language models via next-segment prediction.arXiv preprint arXiv:2412.07380, 2024

    2024 arXiv

  4. [12]

    When to ensemble: Identifying token-level points for stable and fast llm ensembling.arXiv preprint arXiv:2510.15346, 2025

    Heecheol Yun, Kwangmin Ki, Junghyun Lee, and Eunho Yang. When to ensemble: Identifying token-level points for stable and fast llm ensembling.arXiv preprint arXiv:2510.15346, 2025

    2025

  5. [13]

    Llm-blender: Ensembling large language models with pairwise ranking and generative fusion

    Dongfu Jiang, Xiang Ren, and Bill Yuchen Lin. Llm-blender: Ensembling large language models with pairwise ranking and generative fusion. InProceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 14165–14178, 2023

    2023

  6. [14]

    Ensemble modeling for time series forecasting: an adaptive robust optimization approach.arXiv preprint arXiv:2304.04308, 2023

    Dimitris Bertsimas and Leonard Boussioux. Ensemble modeling for time series forecasting: an adaptive robust optimization approach.arXiv preprint arXiv:2304.04308, 2023

    2023 arXiv

  7. [15]

    Dapo: An open-source llm reinforcement learning system at scale.Advances in Neural Information Processing Systems, 38:113222–113244, 2026

    Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Lingjun Liu, et al. Dapo: An open-source llm reinforcement learning system at scale.Advances in Neural Information Processing Systems, 38:113222–113244, 2026

    2026

  8. [16]

    Understanding r1-zero-like training: A critical perspective

    Zichen Liu, Changyu Chen, Wenjun Li, Penghui Qi, Tianyu Pang, Chao Du, Wee Sun Lee, and Min Lin. Understanding r1-zero-like training: A critical perspective. InSecond Conference on Language Modeling

  9. [17]

    Group sequence policy optimization.arXiv preprint arXiv:2507.18071, 2025

    Chujie Zheng, Shixuan Liu, Mingze Li, Xiong-Hui Chen, Bowen Yu, Chang Gao, Kai Dang, Yuqiong Liu, Rui Men, An Yang, et al. Group sequence policy optimization.arXiv preprint arXiv:2507.18071, 2025

    2025 arXiv

  10. [18]

    Soft adaptive policy optimization.arXiv preprint arXiv:2511.20347, 2025

    Chang Gao, Chujie Zheng, Xiong-Hui Chen, Kai Dang, Shixuan Liu, Bowen Yu, An Yang, Shuai Bai, Jingren Zhou, and Junyang Lin. Soft adaptive policy optimization.arXiv preprint arXiv:2511.20347, 2025

    2025 arXiv

  11. [19]

    Semixer: Semantics enhanced mlp-mixer for multiscale mixing and long-term time series forecasting

    Xu Zhang, Qitong Wang, Peng Wang, and Wei Wang. Semixer: Semantics enhanced mlp-mixer for multiscale mixing and long-term time series forecasting. InProceedings of the ACM Web Conference 2026, pages 5636–5647, 2026

    2026

  12. [20]

    A time series is worth 64 words: Long-term forecasting with transformers.arXiv preprint arXiv:2211.14730, 2022

    Yuqi Nie, Nam H Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam. A time series is worth 64 words: Long-term forecasting with transformers.arXiv preprint arXiv:2211.14730, 2022

    2022 arXiv

  13. [21]

    Timexer: Empowering transformers for time series forecasting with exogenous variables

    Yuxuan Wang, Haixu Wu, Jiaxiang Dong, Guo Qin, Haoran Zhang, Yong Liu, Yunzhong 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 Processing Systems

  14. [22]

    Card: Channel aligned robust blend transformer for time series forecasting

    Tian Zhou, Qingsong Wen, Jinyang Gao, Bolin Ding, Rong Jin, et al. Card: Channel aligned robust blend transformer for time series forecasting. InInternational conference on learning representations, volume 2024, pages 11245–11283, 2024. 12 REATS: LLM Reasoning-based Ensemble L...

    2024

  15. [23]

    Timemixer: Decomposable multiscale mixing for time series forecasting

    Shiyu Wang, Haixu Wu, Xiaoming Shi, Tengge Hu, Huakun Luo, Lintao Ma, James Y Zhang, and JUN ZHOU. Timemixer: Decomposable multiscale mixing for time series forecasting. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  16. [24]

    Moderntcn: A modern pure convolution structure for general time series analysis

    Donghao Luo and Xue Wang. Moderntcn: A modern pure convolution structure for general time series analysis. InThe twelfth international conference on learning representations, pages 1–43, 2024

    2024

  17. [25]

    Are transformers effective for time series forecasting? In Proceedings of the AAAI conference on artificial intelligence, volume 37, pages 11121–11128, 2023

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

    2023

  18. [26]

    Periodicity decoupling framework for long-term series forecasting

    Tao Dai, Beiliang Wu, Peiyuan Liu, Naiqi Li, Jigang Bao, Yong Jiang, and Shu-Tao Xia. Periodicity decoupling framework for long-term series forecasting. InThe twelfth international conference on learning representations, 2024

    2024

  19. [27]

    Moment: a family of open time-series foundation models

    Mononito Goswami, Konrad Szafer, Arjun Choudhry, Yifu Cai, Shuo Li, and Artur Dubrawski. Moment: a family of open time-series foundation models. InProceedings of the 41st International Conference on Machine Learning, pages 16115–16152, 2024

    2024

  20. [28]

    Sundial: A family of highly capable time series foundation models.arXiv preprint arXiv:2502.00816, 2025

    Yong Liu, Guo Qin, Zhiyuan Shi, Zhi Chen, Caiyin Yang, Xiangdong Huang, Jianmin Wang, and Mingsheng Long. Sundial: A family of highly capable time series foundation models.arXiv preprint arXiv:2502.00816, 2025

    2025 arXiv

  21. [29]

    Timer: generative pre-trained transformers are large time series models

    Yong Liu, Haoran Zhang, Chenyu Li, Xiangdong Huang, Jianmin Wang, and Mingsheng Long. Timer: generative pre-trained transformers are large time series models. InProceedings of the 41st International Conference on Machine Learning, pages 32369–32399, 2024

    2024

  22. [30]

    Time-moe: Billion-scale time series foundation models with mixture of experts

    Xiaoming Shi, Shiyu Wang, Yuqi Nie, Dianqi Li, Zhou Ye, Qingsong Wen, and Ming Jin. Time-moe: Billion-scale time series foundation models with mixture of experts. InInternational conference on learning representations, volume 2025, pages 34635–34667, 2025

    2025

  23. [31]

    A decoder-only foundation model for time-series forecasting

    Abhimanyu Das, Weihao Kong, Rajat Sen, and Yichen Zhou. A decoder-only foundation model for time-series forecasting. InForty-first International Conference on Machine Learning, pages 10148–10167. PMLR, 2024

    2024

  24. [32]

    Unified training of universal time series forecasting transformers

    Gerald Woo, Chenghao Liu, Akshat Kumar, Caiming Xiong, Silvio Savarese, and Doyen Sahoo. Unified training of universal time series forecasting transformers. 2024

    2024

  25. [33]

    Timer-xl: Long-context transformers for unified time series forecasting.arXiv preprint arXiv:2410.04803, 2024

    Yong Liu, Guo Qin, Xiangdong Huang, Jianmin Wang, and Mingsheng Long. Timer-xl: Long-context transformers for unified time series forecasting.arXiv preprint arXiv:2410.04803, 2024

    2024 arXiv

  26. [34]

    pronounced upward drift

    Abdul Fatir Ansari, Lorenzo Stella, Caner Turkmen, Xiyuan Zhang, Pedro Mercado, Huibin Shen, Oleksandr Shchur, Syama Syndar Rangapuram, Sebastian Pineda Arango, Shubham Kapoor, Jasper Zschiegner, Danielle C. Maddix, Michael W. Mahoney, Kari Torkkola, Andrew Gordon Wilson, Mich...

    2024

  27. [35]

    Electricity dataset3 collects the electricity consumption (kWh) every 15 minutes of 321 clients from 2012 to 2014

    2012

  28. [36]

    Each sub-dataset offers two versions with varying sampling resolutions (15 minutes and 1 hour)

    ETT datasets4 comprises two sub-datasets, ETT1 and ETT2, collected from two separate counties. Each sub-dataset offers two versions with varying sampling resolutions (15 minutes and 1 hour). ETT dataset includes multiple time series of electrical loads and a single time sequen...

  29. [37]

    Weather dataset5 contains 21 meteorological indicators, such as air temperature, humidity, etc, recorded every 10 minutes for the entirety of 2020

    2020

  30. [38]

    Exchange dataset 6 contains the current exchange of eight countries

  31. [39]

    Traffic records hourly road occupancy rates measured by 862 sensors of the San Francisco Bay area freeways in 2 years. Table 7: Statistics of the eight benchmark datasets.Time pointsdenotes the total number of observations.Splitdenotes train/validation/test sizes.Frequencydeno...

  32. [40]

    Near-oracle sensitivity: r(0)−r(0.01)∈[0.05,0.20] , ensuring the function can distinguish near-optimal candidates without over-sensitivity

  33. [41]

    Mid-range signal: r(0.1)∈[0.20,0.50] , maintaining meaningful learning signal while sufficiently penalizing suboptimal generations

  34. [42]

    dead zones

    Far-range signal: r(0.3)>0.05 , ensuring that poorly-performing generations still receive non-zero reward differences for policy optimization. Figure 8 shows the reward curves under varying k. By checking which k values simultaneously satisfy all three criteria, we identify th...

    2022

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.