Pith. sign in

REVIEW 4 major objections 7 minor 49 references

A model-independent training wrapper that down-weights low-predictability samples during training is claimed to improve time series forecasting and classification across sixteen existing models.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 22:36 UTC pith:RZ5OXTDQ

load-bearing objection The long-term and UCR results make the core idea worth a look, but the short-term Fund evaluation contaminates train and test, so the headline range needs a redo. the 4 major comments →

arxiv 2602.16224 v3 pith:RZ5OXTDQ submitted 2026-02-18 cs.LG

Amortized Predictability-aware Training Framework for Time Series Forecasting and Classification

Xu Zhang , Peng Wang , Yichen Li , Wei Wang This is my paper
classification cs.LG
keywords time series forecastingtime series classificationtime series analysisdeep learning techniquesnoise robust learningloss reweightingpredictability-aware training
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper claims that time series models—both forecasting and classification—can be made more accurate and better generalizing by changing only how each training sample's loss is weighted, not the model itself. The proposed approach sorts samples by their current loss into buckets, treats high-loss samples as low-predictability, and progressively down-weights them while retaining a hierarchical memory of earlier bucket partitions. An auxiliary 'amortization' model reduces the source model's bias in this sorting by passing its own bucket assignments to the source one step later. On eleven forecasting and five classification models, the paper reports consistent error reductions, averaging 2–9.79% on long-term and 3–15.36% on short-term forecasting tasks. If true, any existing time series model could be boosted at training time without any change at inference.

Core claim

The paper's central claim is that a sample's current training loss is a serviceable ordering of its predictability, and that reweighting the loss accordingly—down-weighting high-loss (low-predictability) samples and up-weighting low-loss ones—improves convergence and generalization. The hierarchical predictability-aware loss (HPL) implements this by partitioning each batch into K loss-ordered buckets, assigning decreasing weights to higher-loss buckets, and, as training stages advance, averaging over multiple bucket groups so that earlier fine-grained partitions are not discarded. The amortization model is a second network that computes its own bucket assignments and feeds them to the source

What carries the argument

The load-bearing mechanism is HPL, the hierarchical predictability-aware loss: a bucketing scheme that orders training samples by loss magnitude, assigns monotonically decreasing weights to higher-loss buckets, and, at each training stage, computes the loss as an average over several bucket groups from previous stages together with the current one. The amortization model—a second network that supplies the source model with its own bucket assignments with a one-step delay—complements HPL by absorbing the source's bias in estimating predictability.

Load-bearing premise

The method assumes that a sample's current loss value is a faithful ordering of its true predictability, even after reweighting modifies those losses; if overfitting or the reweighting itself corrupts that ordering, HPL could systematically underweight informative samples.

What would settle it

One concrete test: on a synthetic dataset where each sample's noise level is known exactly, train the same model with APTF, with random bucket assignments using identical weights, and with ordinary training. If random buckets match APTF's test-loss gains, then loss-based ordering is not the active ingredient; if APTF hurts on clean datasets without low-predictability samples, the down-weighting is harming rather than helping.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the claim holds, any existing TSF or TSC model can be improved by wrapping its loss with HPL, with no change to inference or architecture.
  • The gains span very different architectures—Transformers, CNNs, and linear models—so the mechanism is orthogonal to model design.
  • HPL alone yields most of the gain; the amortization model adds a further 1.4–2.1% reduction, so the core effect is the loss reweighting.
  • The paper reports flatter loss landscapes after training with APTF, meaning the improved test accuracy is accompanied by better generalization, not just better fit.

Where Pith is reading between the lines

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

  • Because the ordering signal is just the loss, the same idea could be transferred to other sequence tasks (speech, event logs) and to anomaly detection, where predictability is a natural target.
  • A direct test of the causal mechanism would be to compare APTF against random bucket assignments with the same weights; if random buckets perform as well, the loss-based ordering is not what drives the gains.
  • The paper's own appendix notes that full per-dataset standard deviations and per-horizon results are omitted, so the consistency claim is not fully checkable from the text alone.
  • The paper also states its default hyperparameters are not the tuned optimum (e.g., epoch interval 2 vs. a better 3–4), so the reported gains likely underestimate the method's ceiling.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes the Amortized Predictability-aware Training Framework (APTF), a training-time wrapper for time series forecasting (TSF) and classification (TSC). APTF identifies 'low-predictability' samples as those with high loss under the model being trained, groups them into loss-ordered buckets (Hierarchical Predictability-aware Loss, HPL), assigns decreasing weights to lower-predictability buckets, and gradually expands penalization across training stages to address the 'predictability evolution' issue. A second 'amortization model' is added to mutualize predictability estimates and reduce bias. The paper claims consistent accuracy improvements on eleven TSF baselines (2%–9.79% on long-term, 3%–15.36% on short-term tasks) and five TSC baselines on 128 UCR datasets, supported by ablations, comparisons to Co-teaching, WaveBound, ARL, and SPL, loss-landscape analysis, and released code.

Significance. If the empirical claims hold, APTF is a practically useful model-agnostic training wrapper: it changes only the loss, not the inference procedure, and it is tested across a broad model zoo (eleven TSF and five TSC models) and 128 UCR datasets. The paper has several concrete strengths: public code; standard 7:1:2 splits for long-term TSF and original splits for UCR TSC; win/tie summaries over 128 datasets; ablations isolating HPL and the amortization model (Tables 4–5); hyperparameter sensitivity analysis; and a flatter-loss-landscape argument for generalization. The central weakness is the short-term Fund evaluation in §4.1, which as written merges the training, validation, and test sets; Table 1 and the abstract's short-term improvement range therefore cannot be read as evidence of generalization. The long-term TSF and TSC pillars remain independent evidence, but they do not rescue the short-term claims as reported. The verdict is therefore conditional on a clean re-evaluation and on fixing the internal inconsistencies in Algorithm 2 and Table 3.

major comments (4)
  1. [§4.1, Table 1] The short-term TSF evaluation protocol is invalid as written: 'In short-term forecasting, we merged the training, validation, and test sets across all fund datasets for model training and evaluation because they share similar TS patterns.' Taken literally, the test partitions were used in training, so the WMAPE improvements in Table 1 (e.g., TimeMixer Fund1 87.91 vs 89.96; Scaleformer Fund2 96.74 vs 109.01) are not evidence of generalization. If a pooled-split design was intended, the held-out split is nowhere defined, and the sentence must be rewritten. Since §4.2.1 and the abstract's 3%–15.36% short-term range rest on these numbers, this is a load-bearing flaw. Please re-run the Fund experiments with a proper held-out split and report per-seed standard deviations, or remove the short-term claims.
  2. [Algorithm 2] Algorithm 2 as printed cannot implement the described method. The loop 'for i = 1 to S' returns unconditionally in both branches ('return L, eL' at i=1 and 'return L/G, eL/G' at i>1), so the function exits during the first stage and the hierarchical multi-bucket-group computation for later stages—the core of HPL in §3.1.4 and Figure 2(c)—is unreachable. The listing implements only a single-bucket-group loss and contradicts the surrounding text. Please correct the pseudocode: remove the premature returns, make the stage loop or per-stage invocation explicit, define how bucket groups {B_j} and weight groups {W_j} are constructed and expanded across stages, and specify how the amortization model's bucket groups (E-hat, W-hat) are computed and trained; the one-step-delay mechanism in §3.2 is currently not formally specified.
  3. [§3.1.1–§3.1.3] The core identification mechanism is circular by construction: 'low predictability' is defined as high loss of the model being trained, and that same loss is then reweighted to produce the next update. No independent predictability measure is used; the claim that loss values 'reliably reflect sample predictability in early training' (§1) is asserted rather than demonstrated, and the reweighting alters the loss ordering it relies on. This is not fatal to the paper's main empirical claim—external test accuracy is the evidence—but it is a correctness risk and currently unsupported. A concrete fix: on a subset of datasets, report the stability of bucket assignments across seeds and across the source vs. amortization model, or correlate the identified low-predictability samples with an external noisiness measure. The per-batch bucketing (N = batch size, §3.1.1) also makes the ordering noisy a
  4. [Table 3] Table 3's Win/Tie statistics are internally inconsistent. The OSCNN row reports 165 wins and 40 ties on 128 UCR datasets, which is impossible, and no other row's win+tie sum equals 128 (row sums range from 102 to 111), so the number of losses is undefined; the Average row (75/31) does not reconcile either. Either the column meanings or the counts are wrong. Because the TSC claim is that APTF 'consistently' improves five baselines, the counts are part of the evidence and must be corrected: report win/tie/loss numbers that sum to 128 for each model.
minor comments (7)
  1. [§3.1.1] Typo: 'for example, in the in the forecasting task'.
  2. [§4.1 vs Table 3] Naming inconsistency: the TSC baseline is called FCN in §4.1 but FCNet in Table 3.
  3. [§4.1, Tables 1–2] The text states that 'average metrics and standard deviations are reported' for four seeds, but Tables 1–2 and most appendix tables omit standard deviations; the same paragraph defers them to a 'full version paper.' Please either include the variances or state explicitly which tables are complete, since several gains are small (e.g., NLinear ETTh2: 0.380 vs 0.383 in Table 2).
  4. [§4.5] The hyperparameter sensitivity discussion admits that the settings used in the main experiments are not optimal: 'the maximum number of buckets in the experiments is set to 9, which is not optimal' and 'the better epoch interval ε is 3 or 4, while used in the paper is 2.' This raises a question of whether the reported improvements are representative; please clarify whether these suboptimal choices apply evenly to baselines and APTF, and reconcile the text with Figure 4, which shows better performance at K>9.
  5. [Table 6] The WaveBound comparison on Fund datasets yields extreme WMAPE values (e.g., TimeMixer-WaveBound 206.9 vs 89.96 baseline), which suggests the baseline may not have been tuned; since the Fund evaluation is also subject to the merged-set problem (see main comment 1), the comparison is hard to interpret as reported.
  6. [§1, Related Work] The 'first framework' claim would benefit from a qualifier ('to our knowledge') and a positioning paragraph against the broader noise-robust and sample-reweighting literature beyond Co-teaching, self-paced learning, ARL, and WaveBound.
  7. [Abstract, §4.2] The abstract reports improvements over 'eleven TSF and five TSC models,' but Tables 1–2 cover eight TSF baselines and Tables 9–10 cover three additional ones; aggregating these into the abstract ranges would be clearer if the per-model improvement used to compute the 2%–9.79% and 3%–15.36% ranges were reproduced in a summary table.

Circularity Check

1 steps flagged

Short-term TSF prediction claims are contaminated because the paper merges test sets into training, reducing that evaluation pillar to an in-sample comparison.

specific steps
  1. fitted input called prediction [Section 4.1 (Datasets for short-term TSF); Tables 1 and 9; claimed in §4.2.1 and the abstract]
    "In short-term forecasting, we merged the training, validation, and test sets across all fund datasets for model training and evaluation because they share similar TS patterns."

    The central claim's short-term TSF component is quantified by WMAPE reductions in Table 1 and Appendix Table 9 on Fund1–Fund3. By the paper's own sentence, the test partitions of these datasets were merged into the training set used to fit the models. Therefore the 'test' WMAPE values are computed on data that the model saw during training, making the reported 'reductions in prediction errors' in-sample fitting outcomes rather than out-of-sample predictions. The headline '3%–15.36% on short-term TSF tasks' is thus forced by construction: minimizing a training loss on data that includes the evaluation set can only show a favorable comparison on that same set; it is not evidence of generalization.

full rationale

One clear circular-by-construction element is present: the short-term fund-sales benchmark merges test sets into training. That contaminates the short-term pillar of the main claim. The rest of the evidence is largely independent: long-term TSF uses standard 7:1:2 splits, and UCR TSC uses the original fixed train/test partitions; both are external benchmarks with standard protocols. APTF's loss-bucket procedure does define 'low predictability' as high loss of the model being trained and then reweights that same loss, but this is an operational heuristic, not a derivation whose output is identical to its input, and the paper's strongest evidence is external test accuracy rather than the internal predictability estimates. Self-citations ([41]–[44]) are domain applications and are not load-bearing: no uniqueness theorem or ansatz is imported from the authors' own prior work. Missing standard deviations and the admission in §4.5 that chosen hyperparameters are not optimal weaken reporting but are not circularity. Overall, one evaluation pillar reduces by construction, so a partial-circularity score is warranted; the long-term and TSC results remain independent support.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 2 invented entities

The central empirical claim rests on several hand-set hyperparameters and untested domain assumptions; the paper provides no theory for why these settings should transfer. The amortization model is the least specified component, and its benefit is asserted via ablation only.

free parameters (5)
  • initial number of buckets K = 9 for TSF; 5 in illustration; 2 buckets for TSC
    Chosen by hand; §4.5 admits 9 is not optimal and that performance keeps improving up to 15 buckets.
  • training stage interval ε = 2 for TSF, 75 for TSC
    Manual; §4.5 says better TSF values are 3 or 4.
  • loss weight schedule = first bucket weight 1; last bucket half of previous; intermediate decrease evenly by 1/(K-1); largest weight removed eac
    Manual construction; no theoretical justification given.
  • low-predictability bucket weights for TSC = 1 and 0.1 or 0.01
    A.1.1 lists weights 1 and 0.1/0.01; no selection criterion is provided.
  • penalty range expansion rate for TSC = 0.025 per stage
    Manual; §4.5 says this setting is good overall.
axioms (5)
  • domain assumption Loss magnitude is a valid proxy for sample predictability.
    Introduced in §3.1.1: 'Since loss values are effective indicators of noisy samples [14], we use high-loss values to identify low-predictability samples in our setting.'
  • domain assumption Deep networks memorize clean samples first and adapt to noisy samples later (predictability evolution).
    Invoked in §3.1.3 and §3.1.4; supported only by citation [1].
  • domain assumption Co-training style mutual estimation reduces predictability estimation bias.
    §3.2 cites [4]; no derivation or standalone evidence is provided.
  • domain assumption Flatter loss landscape implies better generalization.
    Used in §4.3.3 and Figure 3; citations [22,28] are given but no quantitative link is established.
  • ad hoc to paper The amortization model's one-step-delayed mutual loss computation is a valid approximation of source-model bias.
    Algorithm 2 and §3.2 describe this mechanism without formal justification; the algorithm pseudocode is internally inconsistent.
invented entities (2)
  • Amortization model no independent evidence
    purpose: Second model to estimate sample predictability, absorb source model bias, and provide one-step-delayed bucket assignments.
    No independent falsifiable handle; its contribution is only shown through within-paper ablations and it is not specified enough to evaluate outside the provided code.
  • Hierarchical bucket groups no independent evidence
    purpose: Data structure for progressively penalizing low-predictability samples while averaging previous partitions.
    Method component, not an independently testable entity.

pith-pipeline@v1.3.0-alltime-deepseek · 20027 in / 15596 out tokens · 140856 ms · 2026-08-02T22:36:09.463399+00:00 · methodology

0 comments
read the original abstract

Time series data are prone to noise in various domains, and training samples may contain low-predictability patterns that deviate from the normal data distribution, leading to training instability or convergence to poor local minima. Therefore, mitigating the adverse effects of low-predictability samples is crucial for time series analysis tasks such as time series forecasting (TSF) and time series classification (TSC). While many deep learning models have achieved promising performance, few consider how to identify and penalize low-predictability samples to improve model performance from the training perspective. To fill this gap, we propose a general Amortized Predictability-aware Training Framework (APTF) for both TSF and TSC. APTF introduces two key designs that enable the model to focus on high-predictability samples while still learning appropriately from low-predictability ones: (i) a Hierarchical Predictability-aware Loss (HPL) that dynamically identifies low-predictability samples and progressively expands their loss penalty as training evolves, and (ii) an amortization model that mitigates predictability estimation errors caused by model bias, further enhancing HPL's effectiveness. The code is available at https://github.com/Meteor-Stars/APTF.

Figures

Figures reproduced from arXiv: 2602.16224 by Peng Wang, Wei Wang, Xu Zhang, Yichen Li.

Figure 2
Figure 2. Figure 2: HPL adopts a hierarchical bucketing strategy to penalize [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: The proposed general training framework APTF for time series forecasting and classification. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of different bucketing strategies for computing the predictability-aware loss. (a) A fixed number of [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Loss landscapes of Autoformer with and without [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Hyperparameter sensitivity analysis. stages, similar to Co-teaching, which may lead to underfitting as the model fails to learn sufficiently from the knowledge contained in low-predictability samples. The poor performance of ARL indicates that relying solely on adaptive parameters may not impose suffi￾cient constraints on low-predictability samples, thereby impairing model convergence [PITH_FULL_IMAGE:fig… view at source ↗
Figure 5
Figure 5. Figure 5: Time series visualization of Fund sales dataset. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualizations of forecasting future 10-time steps on Fund1 and Fund2 datasets with the original baseline NHiTS (left) [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Full results of loss landscapes on different baselines with the Fund1 dataset. The flatter the bottom (the larger the [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 5 linked inside Pith

  1. [1]

    Devansh Arpit, Stanisław Jastrzębski, Nicolas Ballas, David Krueger, Emmanuel Bengio, Maxinder S Kanwal, Tegan Maharaj, Asja Fischer, Aaron Courville, Yoshua Bengio, et al. 2017. A closer look at memorization in deep networks. In International conference on machine learning. PMLR, 233–242

    2017

  2. [2]

    Maria-Florina Balcan, Alina Beygelzimer, and John Langford. 2006. Agnostic active learning. InProceedings of the 23rd international conference on Machine learning. 65–72

    2006

  3. [3]

    Jonathan T Barron. 2019. A general and adaptive robust loss function. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition. 4331–4339

    2019

  4. [4]

    Avrim Blum and Tom Mitchell. 1998. Combining labeled and unlabeled data with co-training. InProceedings of the eleventh annual conference on Computational learning theory. 92–100

    1998

  5. [5]

    C Challu, KG Olivares, BN Oreshkin, F Garza, M Mergenthaler, and A Dubrawski

  6. [6]

    Si-An Chen, Chun-Liang Li, Nate Yoder, Sercan O Arik, and Tomas Pfister. 2023. Tsmixer: An all-mlp architecture for time series forecasting.arXiv preprint arXiv:2303.06053(2023)

    Pith/arXiv arXiv 2023

  7. [7]

    Mingyue Cheng, Qi Liu, Zhiding Liu, Zhi Li, Yucong Luo, and Enhong Chen

  8. [8]

    Kyunghyun Cho, Bart Van Merriënboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning phrase representations using RNN encoder-decoder for statistical machine translation. arXiv preprint arXiv:1406.1078(2014)

    Pith/arXiv arXiv 2014

  9. [9]

    Youngin Cho, Daejin Kim, Dongmin Kim, Mohammad Azam Khan, and Jaegul Choo. 2022. WaveBound: dynamic error bounds for stable time series forecasting. Advances in Neural Information Processing Systems35 (2022), 19579–19591

    2022

  10. [10]

    Abhimanyu Das, Weihao Kong, Andrew Leach, Shaan Mathur, Rajat Sen, and Rose Yu. 2023. Long-term Forecasting with TiDE: Time-series Dense Encoder.Trans. Mach. Learn. Res.2023 (2023). https://openreview.net/forum?id=pCbC3aQB5W

    2023

  11. [11]

    Francis X Diebold and Georg Strasser. 2013. On the correlation structure of microstructure noise: A financial economic approach.Review of Economic Studies 80, 4 (2013), 1304–1337

    2013

  12. [12]

    Germain Forestier, François Petitjean, Pavel Senin, Fabien Despinoy, Arnaud Huaulmé, Hassan Ismail Fawaz, Jonathan Weber, Lhassane Idoumghar, Pierre- Alain Muller, and Pierre Jannin. 2018. Surgical motion analysis using discrimina- tive interpretable patterns.Artificial intelligence in medicine91 (2018), 3–11

    2018

  13. [13]

    Yoav Freund, Robert Schapire, and Naoki Abe. 1999. A short introduction to boosting.Journal-Japanese Society For Artificial Intelligence14, 771-780 (1999), 1612

    1999

  14. [14]

    Bo Han, Quanming Yao, Xingrui Yu, Gang Niu, Miao Xu, Weihua Hu, Ivor Tsang, and Masashi Sugiyama. 2018. Co-teaching: Robust training of deep neural net- works with extremely noisy labels.Advances in neural information processing systems31 (2018)

    2018

  15. [15]

    Tao Huang, Pengfei Chen, and Ruipeng Li. 2022. A semi-supervised vae based ac- tive anomaly detection framework in multivariate time series for online systems. InProceedings of the ACM Web Conference 2022. 1797–1806

    2022

  16. [16]

    Hassan Ismail Fawaz, Germain Forestier, Jonathan Weber, Lhassane Idoumghar, and Pierre-Alain Muller. 2019. Deep learning for time series classification: a review.Data Mining and Knowledge Discovery(Jul 2019), 917–963. doi:10.1007/ s10618-019-00619-1

    2019

  17. [17]

    Hassan Ismail Fawaz, Benjamin Lucas, Germain Forestier, Charlotte Pelletier, Daniel F Schmidt, Jonathan Weber, Geoffrey I Webb, Lhassane Idoumghar, Pierre- Alain Muller, and François Petitjean. 2020. Inceptiontime: Finding alexnet for time series classification.Data Mining and Knowledge Discovery34, 6 (2020), 1936–1962

    2020

  18. [18]

    Lu Jiang, Zhengyuan Zhou, Thomas Leung, Li-Jia Li, and Li Fei-Fei. 2018. Mentor- net: Learning data-driven curriculum for very deep neural networks on corrupted labels. InInternational conference on machine learning. PMLR, 2304–2313

    2018

  19. [19]

    Michael I Jordan. 1997. Serial order: A parallel distributed processing approach. InAdvances in psychology. Vol. 121. Elsevier, 471–495

    1997

  20. [20]

    D Kim, J Park, J Lee, and H Kim. 2024. Are Self-Attentions Effective for Time Series Forecasting?. In38th Conference on Neural Information Processing Systems (NeurIPS 2024)

    2024

  21. [21]

    M Kumar, Benjamin Packer, and Daphne Koller. 2010. Self-paced learning for latent variable models.Advances in neural information processing systems23 (2010)

    2010

  22. [22]

    Hao Li, Zheng Xu, Gavin Taylor, Christoph Studer, and Tom Goldstein. 2018. Visualizing the loss landscape of neural nets.Advances in neural information processing systems31 (2018)

    2018

  23. [23]

    Yong Liu, Haixu Wu, Jianmin Wang, and Mingsheng Long. 2022. Non-stationary transformers: Exploring the stationarity in time series forecasting.Advances in Neural Information Processing Systems35 (2022), 9881–9893

    2022

  24. [24]

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

    2024

  25. [25]

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

  26. [26]

    P Nunes, J Santos, and E Rocha. 2023. Challenges in predictive maintenance–A review.CIRP Journal of Manufacturing Science and Technology40 (2023), 53–67

    2023

  27. [27]

    Diptangshu Pandit, Li Zhang, Chengyu Liu, Nauman Aslam, Samiran Chattopad- hyay, and Chee Peng Lim. 2017. Noise reduction in ECG signals using wavelet transform and dynamic thresholding.Emerging trends in neuro engineering and neural computation(2017), 193–206

    2017

  28. [28]

    Namuk Park and Songkuk Kim. 2022. How Do Vision Transformers Work?. In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022. OpenReview.net. https://openreview.net/forum?id= D78Go4hVcxO

    2022

  29. [29]

    Pritam Sarkar and Ali Etemad. 2020. Self-supervised ECG representation learning for emotion recognition.IEEE Transactions on Affective Computing13, 3 (2020), 1541–1554

    2020

  30. [30]

    Rajat Sen, Hsiang-Fu Yu, and Inderjit S Dhillon. 2019. Think globally, act locally: A deep neural network approach to high-dimensional time series forecasting. Advances in neural information processing systems32 (2019)

    2019

  31. [31]

    Amin Shabani, Amir Abdi, Lili Meng, and Tristan Sylvain. 2022. Scaleformer: iterative multi-scale refining transformers for time series forecasting.arXiv preprint arXiv:2206.04038(2022)

    Pith/arXiv arXiv 2022

  32. [32]

    Wensi Tang, Guodong Long, Lu Liu, Tianyi Zhou, Michael Blumenstein, and Jing Jiang. 2022. Omni-Scale CNNs: a simple and effective kernel size configuration for time series classification. InThe Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022. OpenReview.net. https: //openreview.net/forum?id=PDYs7Z2XFGv

    2022

  33. [33]

    Wensi Tang, Guodong Long, Lu Liu, Tianyi Zhou, Jing Jiang, and Michael Blu- menstein. 2020. Rethinking 1d-cnn for time series classification: A stronger baseline.arXiv preprint arXiv:2002.10061(2020), 1–7

    Pith/arXiv arXiv 2020

  34. [34]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need.Advances in neural information processing systems30 (2017)

    2017

  35. [35]

    Shiyu Wang, Haixu Wu, Xiaoming Shi, Tengge Hu, Huakun Luo, Lintao Ma, James Y Zhang, and Jun Zhou. 2024. TimeMixer: Decomposable Multiscale Mixing for Time Series Forecasting. InThe Twelfth International Conference on Learning Representations

    2024

  36. [36]

    Yuxuan Wang, Haixu Wu, Jiaxiang Dong, Guo Qin, Haoran Zhang, Yong Liu, Yunzhong Qiu, Jianmin Wang, and Mingsheng Long. [n. d.]. TimeXer: Empower- ing Transformers for Time Series Forecasting with Exogenous Variables. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems

  37. [37]

    Zhiguang Wang, Weizhong Yan, and Tim Oates. 2017. Time series classification from scratch with deep neural networks: A strong baseline. In2017 International joint conference on neural networks (IJCNN). IEEE, 1578–1585

    2017

  38. [38]

    Haixu Wu, Jiehui Xu, Jianmin Wang, and Mingsheng Long. 2021. Autoformer: De- composition transformers with auto-correlation for long-term series forecasting. Advances in Neural Information Processing Systems34 (2021), 22419–22430

    2021

  39. [39]

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

    2023

  40. [40]

    Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals

  41. [41]

    Xu Zhang, Zhengang Huang, Yunzhi Wu, Xun Lu, Erpeng Qi, Yunkai Chen, Zhongya Xue, Peng Wang, and Wei Wang. 2024. Self-Adaptive Scale Handling for Forecasting Time Series with Scale Heterogeneity. InICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 7485–7489

    2024

  42. [42]

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

    2025

  43. [43]

    Xu Zhang, Peng Wang, Chen Wang, Zhe Xu, Xiaohua Nie, and Wei Wang. 2025. Global Feature Enhancing and Fusion Framework for Strain Gauge Status Recog- nition. InCompanion Proceedings of the ACM on Web Conference 2025. 611–620

    2025

  44. [44]

    Xu Zhang, Qitong Wang, Peng Wang, and Wei Wang. 2025. A Lightweight Sparse Interaction Network for Time Series Forecasting. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 13304–13312

    2025

  45. [45]

    Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang. 2021. Informer: Beyond efficient transformer for long se- quence time-series forecasting. InProceedings of the AAAI conference on artificial intelligence, Vol. 35. 11106–11115

    2021

  46. [46]

    Tian Zhou, Ziqing Ma, Qingsong Wen, Liang Sun, Tao Yao, Wotao Yin, Rong Jin, et al. 2022. Film: Frequency improved legendre memory model for long-term time series forecasting.Advances in Neural Information Processing Systems35 (2022), 12677–12690. WWW ’26, April 13–17, 2026, Dubai, United Arab Emirates Xu Zhang, PengWang, Yichen Li, and Wei Wang A Appendi...

    2022

  47. [2021]

    Understanding deep learning (still) requires rethinking generalization. Commun. ACM64, 3 (2021), 107–115

    2021

  48. [2022]

    N-hits: Neural hierarchical interpolation for time series forecasting. arXiv. arXiv preprint arXiv:2201.12886(2022)

    Pith/arXiv arXiv 2022

  49. [2023]

    InProceedings of the ACM Web Conference 2023

    Formertime: Hierarchical multi-scale representations for multivariate time series classification. InProceedings of the ACM Web Conference 2023. 1437–1445

    2023