Pith. sign in

REVIEW 4 major objections 6 minor 52 references

Act Now: A Novel Online Forecasting Framework for Large-Scale Streaming Data

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read Act-Now claims that large-scale streaming forecasting can be made causal and accurate by updating with consistent pseudo-labels, and reports large error reductions over continual-learning baselines.

desk verdict Useful online-forecasting framework, but the paper's headline numbers and SOTA claim don't survive contact with its own Table I. read the letter →

arxiv 2412.00108 v1 pith:XPHEVMZH submitted 2024-11-28 cs.LG

classification cs.LG
keywords onlineforecastingstreamingdataconceptdriftrandomsubgraphsamplinglabeldecompositioncontinuallearningspatio-temporalinformationleakage
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 tries to establish that online forecasting of very large streaming data can be done both causally and accurately. Existing online methods update models with future labels, leaking information; when that leakage is removed, accuracy collapses. The proposed Act-Now framework instead keeps the model learning immediately with partial labels plus consistent pseudo-labels, and separately uses complete labels from earlier moments, so nothing future is seen. On three large real-world city-scale datasets, the label-decomposition model Lade reports lower MSE than the continual-learning baselines, with average 28.4% and 19.5% relative reductions on the CBS and Milano datasets. A sympathetic reader would care because this is a path to GPU-feasible, causality-respecting online forecasting at city scale.

What carries the argument

The central mechanism is a split-stream update scheme. A Fast Stream Buffer (FSB) updates the model immediately using partial true labels and 'consistent pseudo-labels'—the prediction made from recent inputs is treated as a target for the older prediction via Eq. (2), on the assumption (Proposition 1) that newer predictions are better. A Slow Stream Buffer (SSB) updates the model in parallel on other GPUs with complete labels from earlier time steps, and the validation set is itself updated online. To resist concept drift, the Lade model decomposes each target into a mean, variance, and normalized residual flow (mean-variance decomposition), predicts each component with separate predictors, and recombines them through a learnable combiner; the statistical flow and normalization flow are trained by separate optimizers. Random Subgraph Sampling (RSS) cuts the large graph into random subgraphs, with Theorem 1 stating that the sampled subgraph's aggregated features are an unbiased estimate of the full graph's features.

What would settle it

Take a stream with sudden, repeated regime shifts, and compare Act-Now with FSB enabled versus disabled: if MSE and pseudo-label error rise together when drift is abrupt, Proposition 1 fails. More directly, at each FSB update, check whether the newer prediction's error is actually lower than the older prediction's on the eventual ground truth; if it is not, the pseudo-label target is misleading.

Watch

Extended reading notes

Core claim

The paper's central claim is that its Act-Now framework—random subgraph sampling, fast and slow stream buffers, the Lade label-decomposition model, and online updates on the validation set—makes large-scale streaming forecasting both leakage-free and state of the art. On the CBS, Milano, and C2TM datasets, Lade is reported to achieve the lowest MSE/MAE across all prediction lengths compared with Experience Replay, MIR, FSNet, OnlineTCN, DER++, and offline training. The reported headline numbers are a 28.4% average relative cumulative MSE reduction over ER and MIR on CBS and 19.5% on Milano. The framework also claims versatility: wrapping other forecasting models in it improves them, for example Informer by 56% and 41% at horizons 24 and 48.

Load-bearing premise

The whole fast-update loop assumes that a prediction made from recent data is more accurate than one made from older data; if recent predictions are biased or noisy—say, right after a sudden change—the model learns from its own mistakes and the gains vanish.

Editorial extensions

If this is right

  • Removing information leakage no longer has to cost accuracy: immediate updates can come from partial labels and consistent pseudo-labels while full labels arrive later.
  • Large graph-structured streams can be trained on a single GPU by repeatedly sampling subgraphs, since RSS is claimed to be an unbiased estimator of full-graph features.
  • Validation data can be folded into online training, so the model keeps learning continuously instead of freezing after model selection.
  • The framework is modular: other forecasting backbones plugged into Act-Now improve, often by double-digit percentages, so the benefits are not tied to Lade alone.
  • Lade's separate statistical and normalization flows give faster training and inference than transformer baselines while keeping lower MSE.

Reading between the lines

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

  • If consistent pseudo-labels work because nearby predictions are more accurate, the same trick could apply to any delayed-label setting—recommender systems, sensor fusion, or reinforcement learning—where an immediate target is needed before ground truth arrives.
  • The claimed gains likely depend on stream smoothness: under abrupt regime changes, newer predictions may be biased, so a change-point detector or confidence gate on pseudo-labels would be a natural extension the paper does not test.
  • RSS's unbiasedness result suggests the approach could scale beyond the tested tens of thousands of nodes to much larger networks, or be combined with variational node sampling to control which parts of the graph get updated most often.
  • Lade's variance component is trained as a target, not just a statistic, which may give prediction-interval estimates for free; the paper notes this at the end but does not quantify coverage.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes Act-Now, an online forecasting framework for large-scale streaming data. It introduces Random Subgraph Sampling (RSS) to partition large graphs into GPU-sized subgraphs, a Fast Stream Buffer (FSB) that updates the model with partial labels and consistent pseudo-labels, a Slow Stream Buffer (SSB) that updates the model in parallel with earlier full labels, online updates on the validation set, and a Label Decomposition model (Lade) with statistical and normalization flows. The central claims are that Lade achieves state-of-the-art performance across all datasets and prediction lengths and that the framework yields average MSE improvements of 28.4% and 19.5% over existing methods. Experiments are reported on the CBS, Milano, and C2TM datasets, and a GitHub repository is provided for reproducibility.

Significance. If the claims held, Act-Now would be a useful contribution: it addresses a genuine issue in online forecasting (information leakage in model updates), proposes a GPU-feasible sampling strategy for large graph-structured streams, and provides an ablation study plus an open-source implementation. The Lade architecture and the dual-buffer design are interesting. However, the headline quantitative claims are not supported by the paper's own Table I, the consistent-SOTA statement is contradicted by the C2TM input-8-predict-4 result, and the FSB pseudo-label mechanism rests on an unproved superiority assumption. The framework is potentially valuable, but the evidence as presented requires correction and more careful evaluation before the stated conclusions can be accepted.

major comments (4)
  1. [Abstract; Section VI-D; Table I] The claim that Lade 'consistently achieves state-of-the-art (SOTA) performance across all datasets and prediction length settings' is contradicted by Table I: for C2TM with input length 8 and prediction length 4, Lade reports MSE 1.984, while Offline and OnlineTCN report 1.974, and ER, MIR, and DER++ report 1.976. The text partly acknowledges this in Section VI-D ('Lade performs worse than the offline model on the input-8-output-4 setup'), but the unqualified SOTA statement appears in the same section and in the abstract. The SOTA claim must be corrected to reflect the actual configuration.
  2. [Abstract; Section VI-D; Table I] The abstract and Section VI-D report average improvements of 28.4% and 19.5% 'compared to the advanced ER and MIR' on CBS and Milano, respectively. The numbers in Table I do not support this aggregation. For CBS, the summed MSE over prediction lengths 24, 48, and 72 is 9.105 for Lade, 10.205 for ER (a 10.8% reduction), and 10.136 for MIR (10.2%). For Milano, the summed MSE is 6.297 for Lade, 7.843 for ER (19.7%, close to the stated 19.5%), but 6.699 for MIR (6.0%). No standard aggregation over the reported rows yields 28.4% for CBS. The authors must either state the exact formula used (e.g., which baselines are averaged, whether horizons are pooled) or revise the headline claims.
  3. [Section IV-D, Eq. (2)] The consistent pseudo-label loss trains the model to match its own newer predictions, so the only stated justification is Proposition 1, which asserts that newer-input predictions are superior without proof or measurement. The distance metric in Eq. (2) is also not defined. Because this loss is part of the FSB contribution, the FSB ablation results in Fig. 6(b) are difficult to interpret: under abrupt concept drift the pseudo-label could be biased and the loss could reinforce errors. The full-label loss Eq. (10) and SSB provide independent supervision, so the framework is not circular overall, but the value of the FSB component needs a test that compares against partial-label-only updates or injects known drift to check pseudo-label quality.
  4. [Table III vs Table I] Table III reports an offline Lade MSE of 1.841 for Milano at prediction length 24, while Table I reports the offline MSE on the same setting as 2.330, and Table II reports values of 2.268, 2.330, and 2.357 depending on the number of partitions. The online Lade value in Table III (1.802) also differs from Table I (1.797). The paper does not explain these discrepancies, which makes the performance-improvement percentages in Table III hard to verify. Please clarify the experimental setup or use one consistent set of results.
minor comments (6)
  1. [Section VI-F] The text uses 'SSB+FFB+Val' where the second component should be 'FSB'; the acronym FFB is not introduced anywhere.
  2. [Section VI-D] The text uses 'input-8-output-4' while the dataset description uses 'Input-8-predict-4'; please use one notation throughout the paper.
  3. [Appendix B] In the proof of Theorem 1, Eq. (12) writes the sampled neighborhood as N(v) \cup V', but the expectation step uses only N(v); the sampling probability P(u) is not defined in terms of Algorithm 1. Please rewrite the proof with explicit notation for the sampled set and the probability model.
  4. [Eqs. (5)-(6)] The value of the constant epsilon is not given in Section VI-B, and Var(Y', -1) does not specify the axis along which the variance is computed; please state these details.
  5. [Section VI-J, Table IV] The column labeled 'Inference (S/Epoch)' is misleading because inference is not measured per epoch; rename it (for example, 'inference time per pass') and state the batch size.
  6. [Section VI-E] Figure 6(b) reports ablation results without error bars, even though the text says all methods are run at least three times; please add error bars or state why they are omitted.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity in FSB's pseudo-label target; central SOTA claim is independently grounded.

  1. self definitional [Section IV-D, Proposition 1 and Eq. (2)]
    "Proposition 1: (Consistent Pseudo-Labels) For predictions over a future time period, assume that results ˆYnew derived from recent inputs are superior to those ˆYold based on inputs from more distant time points. Therefore, results ˆYnew can be used as consistent pseudo-labels for forecasts ˆYold. ... Lcpl = Distance([Ypart, ˆYnew], ˆYold)"

    The FSB training target is built from the model's own output: the newer prediction Ŷnew acts as the label for the older prediction Ŷold of the same network, with only a partial-label prefix. Minimizing Eq. (2) therefore enforces self-agreement between two outputs of the same model rather than supplying external ground truth for the unlabeled positions. The only support offered is Proposition 1's assumption of newer-prediction superiority, which is asserted, not proved; under concept drift the pseudo-label can be the model's own error. This is a component-level, partial circularity: Eq. (10) and SSB still provide independent full-label supervision, and the main SOTA claim is measured on held-out MSE, so the central result is not forced by this step.

full rationale

The only load-bearing step that is self-referential by construction is the FSB consistent pseudo-label loss (Section IV-D, Eq. (2)), where the target for the older prediction is the newer prediction of the same model. This is a real but partial circularity: it can reinforce the model's own errors if the "newer is better" assumption fails, and the ablation in Fig. 6(b) attributes only a small gain (1.4% on Milano) to FSB. The central empirical claim (Lade SOTA) is supported by independent held-out MSE comparisons against baselines, and the RSS unbiasedness proof (Appendix VII-B) is a self-contained importance-sampling argument. Self-citations ([46] PSLD, [49] Periodformer) are present but not load-bearing: they appear as baselines, not as justification for Act-Now's correctness. The discrepancy between the abstract's 28.4%/19.5% improvements and Table I is a numerical consistency / correctness issue, not circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim relies on an unproven pseudo-label superiority assumption, on a validation-set update protocol that can double-use data, and on an RSS proof that assumes independent sampling not exactly implemented. No invented physical entities are introduced.

free parameters (3)
  • Number of RSS subgraph partitions Npart = 17 (CBS), 25 (Milano), 62 (C2TM)
    Chosen per dataset; controls subgraph size and GPU feasibility. Ablation on Milano shows stability across 10/20/25, but no principled selection rule is given.
  • Stream update frequency Dfreq = not reported for the main results
    Central to the framework's label-availability claim, yet Table I does not state the Dfreq used; Section VI-E ablates 1/2/4/6 only on Milano.
  • Epsilon in Mean-Variance Decomposer = not specified
    Introduced in Eq. (5) to avoid division by zero; the exact value is not given, so exact normalization cannot be replicated.
assumptions (4)
  • ad hoc to paper Newer predictions are superior to older predictions, so the newer prediction can serve as a pseudo-label for the older prediction (Proposition 1).
    Load-bearing for FSB; no proof, no measurement, and can fail under sudden concept drift or noise.
  • domain assumption The validation set can be used for online updates without corrupting model selection or evaluation.
    Section IV-F. If the same validation set also selects hyperparameters or checkpoints, updating on it double-uses the data and can inflate test results.
  • standard math RSS samples nodes independently so the expectation of the indicator equals the sampling probability.
    Appendix VII-B relies on independent sampling, while Algorithm 1 draws a fixed-size random subset; the proof as written does not exactly match the implementation.
  • domain assumption At the current time only partial labels are available; complete labels arrive only after the forecast horizon.
    This causal label-availability model defines the problem and motivates FSB/SSB; it is a legitimate but strong restriction on the evaluation protocol.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Act Now: A Novel Online Forecasting Framework for Large-Scale Streaming Data." pith.science (2026). https://pith.science/paper/XPHEVMZH

@misc{pith2026241200108,
  author       = {Pith},
  title        = {Pith review of: Act Now: A Novel Online Forecasting Framework for Large-Scale Streaming Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XPHEVMZH}},
  note         = {Machine review of arXiv:2412.00108}
}
read the original abstract

In this paper, we find that existing online forecasting methods have the following issues: 1) They do not consider the update frequency of streaming data and directly use labels (future signals) to update the model, leading to information leakage. 2) Eliminating information leakage can exacerbate concept drift and online parameter updates can damage prediction accuracy. 3) Leaving out a validation set cuts off the model's continued learning. 4) Existing GPU devices cannot support online learning of large-scale streaming data. To address the above issues, we propose a novel online learning framework, Act-Now, to improve the online prediction on large-scale streaming data. Firstly, we introduce a Random Subgraph Sampling (RSS) algorithm designed to enable efficient model training. Then, we design a Fast Stream Buffer (FSB) and a Slow Stream Buffer (SSB) to update the model online. FSB updates the model immediately with the consistent pseudo- and partial labels to avoid information leakage. SSB updates the model in parallel using complete labels from earlier times. Further, to address concept drift, we propose a Label Decomposition model (Lade) with statistical and normalization flows. Lade forecasts both the statistical variations and the normalized future values of the data, integrating them through a combiner to produce the final predictions. Finally, we propose to perform online updates on the validation set to ensure the consistency of model learning on streaming data. Extensive experiments demonstrate that the proposed Act-Now framework performs well on large-scale streaming data, with an average 28.4% and 19.5% performance improvement, respectively. Experiments can be reproduced via https://github.com/Anoise/Act-Now.

Figures

Figures reproduced from arXiv: 2412.00108 by the authors.

Figure 1
Figure 1. (a) Information leakage in streaming data forecasting. (b) Removing [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) Removing information leakage exacerbate concept drift, and model updating online may damage forecasting. (b) Leaving out the validation set [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Random Subgraph Sampling (RSS): For large-scale graph-structured data (a), a subgraph is randomly selected at each iteration (b). Through multiple [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Both FSB and SSB are implemented through streaming buffers (bottom of the figures). (a) FSB updates the model online through partial labels and [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: The architecture of Lade, which including three main parts: decomposer [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: (a) Ablation studies on the update frequency of streaming data using the Milano datasets. (b) Ablation studies on the components of Lade using CBS [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Visualization of Lade prediction results utilizing MVD. Prediction cases from the Milano dataset under the input-36-predict-72 setting. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Visualization of the prediction results across multiple models. Prediction cases from the Milano dataset under the input-36-predict-72 setting. [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 22 canonical work pages

  1. [1]

    A variegated look at 5g in the wild: performance, power, and qoe implications,

    A. Narayanan, X. Zhang, R. Zhu, A. Hassan, S. Jin, X. Zhu, X. Zhang, D. Rybkin, Z. Yang, Z. M. Mao et al., “A variegated look at 5g in the wild: performance, power, and qoe implications,” in Proceedings of the 2021 ACM SIGCOMM 2021 Conference , 2021, pp. 610–625

  2. [2]

    Evolution of wireless communication to 6g: Potential applications and research directions,

    M. Z. Asghar, S. A. Memon, and J. H ¨am¨al¨ainen, “Evolution of wireless communication to 6g: Potential applications and research directions,” Sustainability, vol. 14, no. 10, p. 6356, 2022

  3. [3]

    Cell zooming for cost-efficient green cellular networks,

    Z. Niu, Y . Wu, J. Gong, and Z. Yang, “Cell zooming for cost-efficient green cellular networks,” IEEE communications magazine , vol. 48, no. 11, pp. 74–79, 2010

  4. [4]

    The deep learning vision for heterogeneous network traffic control: Proposal, challenges, and future perspective,

    N. Kato, Z. M. Fadlullah, B. Mao, F. Tang, O. Akashi, T. Inoue, and K. Mizutani, “The deep learning vision for heterogeneous network traffic control: Proposal, challenges, and future perspective,” IEEE wireless communications, vol. 24, no. 3, pp. 146–153, 2016

  5. [5]

    Deepcog: Optimizing resource provisioning in network slicing with ai-based capacity forecasting,

    D. Bega, M. Gramaglia, M. Fiore, A. Banchs, and X. Costa-P ´erez, “Deepcog: Optimizing resource provisioning in network slicing with ai-based capacity forecasting,” IEEE Journal on Selected Areas in Communications, vol. 38, no. 2, pp. 361–376, 2020

  6. [6]

    Intelligent 5g: When cellular networks meet artificial intelligence,

    R. Li, Z. Zhao, X. Zhou, G. Ding, Y . Chen, Z. Wang, and H. Zhang, “Intelligent 5g: When cellular networks meet artificial intelligence,” IEEE Wireless communications, vol. 24, no. 5, pp. 175–183, 2017

  7. [7]

    Online learning for time series prediction,

    O. Anava, E. Hazan, S. Mannor, and O. Shamir, “Online learning for time series prediction,” in Conference on learning theory. PMLR, 2013, pp. 172–184

  8. [8]

    Online arima algorithms for time series prediction,

    C. Liu, S. C. Hoi, P. Zhao, and J. Sun, “Online arima algorithms for time series prediction,” in Thirtieth AAAI conference on artificial intelligence, 2016

Show all 52 references
  1. [9]

    Laplace propagation

    A. J. Smola, V . Vishwanathan, and E. Eskin, “Laplace propagation.” in NIPS. Citeseer, 2003, pp. 441–448

  2. [10]

    Online forecasting matrix factorization,

    S. Gultekin and J. Paisley, “Online forecasting matrix factorization,” IEEE Transactions on Signal Processing, vol. 67, no. 5, pp. 1223–1236, 2018

  3. [11]

    Continual learning with bayesian neural networks for non-stationary data,

    R. Kurle, B. Cseke, A. Klushyn, P. Van Der Smagt, and S. G ¨unnemann, “Continual learning with bayesian neural networks for non-stationary data,” in International Conference on Learning Representations , 2019

  4. [12]

    Dynamic local regret for non- convex online forecasting,

    S. Aydore, T. Zhu, and D. P. Foster, “Dynamic local regret for non- convex online forecasting,” Advances in Neural Information Processing Systems, vol. 32, pp. 7982–7991, 2019

  5. [13]

    Cost: Contrastive learning of disentangled seasonal-trend representations for time series forecasting,

    G. Woo, C. Liu, D. Sahoo, A. Kumar, and S. Hoi, “Cost: Contrastive learning of disentangled seasonal-trend representations for time series forecasting,” arXiv preprint arXiv:2202.01575 , 2022

  6. [14]

    Onenet: Enhancing time series forecasting models under concept drift by online ensembling,

    Y .-F. Zhang, Q. Wen, X. Wang, W. Chen, L. Sun, Z. Zhang, L. Wang, R. Jin, and T. Tan, “Onenet: Enhancing time series forecasting models under concept drift by online ensembling,” arXiv preprint arXiv:2309.12659, 2023

  7. [15]

    Learning fast and slow for online time series forecasting,

    Q. Pham, C. Liu, D. Sahoo, and S. Hoi, “Learning fast and slow for online time series forecasting,” in The Eleventh International Conference on Learning Representations , 2023

  8. [16]

    Forecasting at scale,

    S. J. Taylor and B. Letham, “Forecasting at scale,” The American Statistician, vol. 72, no. 1, pp. 37–45, 2018

  9. [17]

    N-beats: Neural basis expansion analysis for interpretable time series forecasting,

    B. N. Oreshkin, D. Carpov, N. Chapados, and Y . Bengio, “N-beats: Neural basis expansion analysis for interpretable time series forecasting,” in International Conference on Learning Representations , 2019

  10. [18]

    Think globally, act locally: A deep neural network approach to high-dimensional time series forecasting,

    R. Sen, H.-F. Yu, and I. S. Dhillon, “Think globally, act locally: A deep neural network approach to high-dimensional time series forecasting,” Advances in neural information processing systems , vol. 32, 2019

  11. [19]

    Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,

    H. Wu, J. Xu, J. Wang, and M. Long, “Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,” in Advances in Neural Information Processing Systems (NeurIPS) , vol. 34, Virtual Conference, 2021, pp. 22 419–22 430

  12. [20]

    FED- former: Frequency enhanced decomposed transformer for long-term series forecasting,

    T. Zhou, Z. Ma, Q. Wen, X. Wang, L. Sun, and R. Jin, “FED- former: Frequency enhanced decomposed transformer for long-term series forecasting,” in Proceedings of the 39th International Conference on Machine Learning (ICML), vol. 162, Baltimore, Maryland, 2022, pp. 27 268–27 286

  13. [21]

    Non-stationary transformers: Exploring the stationarity in time series forecasting,

    Y . Liu, H. Wu, J. Wang, and M. Long, “Non-stationary transformers: Exploring the stationarity in time series forecasting,” Advances in Neural Information Processing Systems , vol. 35, pp. 9881–9893, 2022

  14. [22]

    The problem of concept drift: definitions and related work,

    A. Tsymbal, “The problem of concept drift: definitions and related work,” Computer Science Department, Trinity College Dublin , vol. 106, no. 2, p. 58, 2004

  15. [23]

    Ddg-da: Data distribution generation for predictable concept drift adaptation,

    W. Li, X. Yang, W. Liu, Y . Xia, and J. Bian, “Ddg-da: Data distribution generation for predictable concept drift adaptation,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 36, no. 4, 2022, pp. 4092–4100

  16. [24]

    Generalizing to evolving domains with latent structure-aware sequential autoencoder,

    T. Qin, S. Wang, and H. Li, “Generalizing to evolving domains with latent structure-aware sequential autoencoder,” in International Confer- ence on Machine Learning . PMLR, 2022, pp. 18 062–18 082

  17. [25]

    Online deep learning: learning deep neural networks on the fly,

    D. Sahoo, Q. Pham, J. Lu, and S. C. Hoi, “Online deep learning: learning deep neural networks on the fly,” inProceedings of the 27th International Joint Conference on Artificial Intelligence , 2018, pp. 2660–2666

  18. [26]

    A survey on concept drift adaptation,

    J. Gama, I. ˇZliobait˙e, A. Bifet, M. Pechenizkiy, and A. Bouchachia, “A survey on concept drift adaptation,” ACM computing surveys (CSUR) , vol. 46, no. 4, pp. 1–37, 2014

  19. [27]

    On quadratic penalties in elastic weight consolidation,

    F. Husz ´ar, “On quadratic penalties in elastic weight consolidation,”arXiv preprint arXiv:1712.03847, 2017

  20. [28]

    Reply to husz ´ar: The elastic weight consolidation penalty is empirically valid,

    J. Kirkpatrick, R. Pascanu, N. Rabinowitz, J. Veness, G. Desjardins, A. A. Rusu, K. Milan, J. Quan, T. Ramalho, A. Grabska-Barwinska et al. , “Reply to husz ´ar: The elastic weight consolidation penalty is empirically valid,” Proceedings of the National Academy of Sciences , v...

  21. [29]

    Gradient episodic memory for continual learning,

    D. Lopez-Paz and M. Ranzato, “Gradient episodic memory for continual learning,” Advances in neural information processing systems , vol. 30, pp. 6467–6476, 2017

  22. [30]

    How does a brain build a cognitive code?

    S. Grossberg, “How does a brain build a cognitive code?” Studies of mind and brain , pp. 1–52, 1982

  23. [31]

    Self-improving reactive agents based on reinforcement learn- ing, planning and teaching,

    L.-J. Lin, “Self-improving reactive agents based on reinforcement learn- ing, planning and teaching,” Machine learning, vol. 8, no. 3-4, pp. 293– 321, 1992

  24. [32]

    Learning to learn without forgetting by maximizing transfer and minimizing interference,

    M. Riemer, I. Cases, R. Ajemian, M. Liu, I. Rish, Y . Tu, and G. Tesauro, “Learning to learn without forgetting by maximizing transfer and minimizing interference,” International Conference on Learning Rep- resentations (ICLR), 2019

  25. [33]

    Experi- ence replay for continual learning,

    D. Rolnick, A. Ahuja, J. Schwarz, T. Lillicrap, and G. Wayne, “Experi- ence replay for continual learning,” in Advances in Neural Information Processing Systems, 2019, pp. 348–358

  26. [34]

    Dualnet: Continual learning, fast and slow,

    Q. Pham, C. Liu, and S. Hoi, “Dualnet: Continual learning, fast and slow,” Advances in Neural Information Processing Systems , vol. 34, 2021

  27. [35]

    Learning fast, learning slow: A general continual learning method based on complementary learning sys- tem,

    E. Arani, F. Sarfraz, and B. Zonooz, “Learning fast, learning slow: A general continual learning method based on complementary learning sys- tem,” in International Conference on Learning Representations , 2021

  28. [36]

    Stl: A seasonal-trend decomposition,

    R. B. Cleveland, W. S. Cleveland, J. E. McRae, and I. Terpenning, “Stl: A seasonal-trend decomposition,” J. Off. Stat , vol. 6, no. 1, pp. 3–73, 1990

  29. [37]

    Forecasting time series with complex seasonal patterns using exponential smoothing,

    A. M. De Livera, R. J. Hyndman, and R. D. Snyder, “Forecasting time series with complex seasonal patterns using exponential smoothing,” Journal of the American statistical association , vol. 106, no. 496, pp. 1513–1527, 2011

  30. [38]

    Time series forecasting for nonlinear and non- stationary processes: a review and comparative study,

    C. Cheng, A. Sa-Ngasoongsong, O. Beyca, T. Le, H. Yang, Z. Kong, and S. T. Bukkapatnam, “Time series forecasting for nonlinear and non- stationary processes: a review and comparative study,” Iie Transactions, vol. 47, no. 10, pp. 1053–1071, 2015

  31. [39]

    A multi-source dataset of urban life in the city of milan and the province of trentino,

    G. Barlacchi, M. De Nadai, R. Larcher, A. Casella, C. Chitic, G. Torrisi, F. Antonelli, A. Vespignani, A. Pentland, and B. Lepri, “A multi-source dataset of urban life in the city of milan and the province of trentino,” Scientific data, vol. 2, no. 1, pp. 1–15, 2015

  32. [40]

    Analyzing and modeling spatio-temporal dependence of cellular traffic at city scale,

    Y . J. Xiaming Chen, S. Qiang, W. Hu, and K. Jiang, “Analyzing and modeling spatio-temporal dependence of cellular traffic at city scale,” in Communications (ICC), 2015 IEEE International Conference on , 2015

  33. [41]

    Adam: A method for stochastic optimiza- tion,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimiza- tion,” in International Conference on Learning Representations (ICLR) , Santiago de Cuba, 2015

  34. [42]

    Online convex programming and generalized infinitesi- mal gradient ascent,

    M. Zinkevich, “Online convex programming and generalized infinitesi- mal gradient ascent,” in Proceedings of the 20th international conference on machine learning (icml-03) , 2003, pp. 928–936

  35. [43]

    On tiny episodic memories in continual learning,

    A. Chaudhry, M. Rohrbach, M. Elhoseiny, T. Ajanthan, P. K. Dokania, P. H. Torr, and M. Ranzato, “On tiny episodic memories in continual learning,” arXiv preprint arXiv:1902.10486 , 2019

  36. [44]

    Online continual learning with maximal interfered retrieval,

    R. Aljundi, E. Belilovsky, T. Tuytelaars, L. Charlin, M. Caccia, M. Lin, and L. Page-Caccia, “Online continual learning with maximal interfered retrieval,” Advances in Neural Information Processing Systems , vol. 32, pp. 11 849–11 860, 2019

  37. [45]

    Dark experience for general continual learning: a strong, simple baseline,

    P. Buzzega, M. Boschini, A. Porrello, D. Abati, and S. Calderara, “Dark experience for general continual learning: a strong, simple baseline,” in 34th Conference on Neural Information Processing Systems (NeurIPS 2020), 2020

  38. [46]

    Progressive supervision via label decomposition: An long-term and large-scale wireless traffic forecasting method,

    D. Liang, H. Zhang, D. Yuan, and M. Zhang, “Progressive supervision via label decomposition: An long-term and large-scale wireless traffic forecasting method,” Knowledge-Based Systems , vol. 305, p. 112622, 2024

  39. [47]

    Frequency-domain mlps are more effective learners in time series forecasting,

    K. Yi, Q. Zhang, W. Fan, S. Wang, P. Wang, H. He, N. An, D. Lian, L. Cao, and Z. Niu, “Frequency-domain mlps are more effective learners in time series forecasting,” Advances in Neural Information Processing Systems, vol. 36, 2024

  40. [48]

    Fouriergnn: Rethinking multivariate time series forecast- ing from a pure graph perspective,

    K. Yi, Q. Zhang, W. Fan, H. He, L. Hu, P. Wang, N. An, L. Cao, and Z. Niu, “Fouriergnn: Rethinking multivariate time series forecast- ing from a pure graph perspective,” Advances in Neural Information Processing Systems, vol. 36, 2024

  41. [49]

    Does long- term series forecasting need complex attention and extra long inputs?

    D. Liang, H. Zhang, D. Yuan, X. Ma, D. Li, and M. Zhang, “Does long- term series forecasting need complex attention and extra long inputs?” arXiv preprint arXiv:2306.05035 , 2023

  42. [50]

    Are transformers effective for time series forecasting?

    A. Zeng, M. Chen, L. Zhang, and Q. Xu, “Are transformers effective for time series forecasting?” in Proceedings of the AAAI conference on artificial intelligence, vol. 37, no. 9, 2023, pp. 11 121–11 128

  43. [51]

    Informer: Beyond efficient transformer for long sequence time-series forecasting,

    H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, and W. Zhang, “Informer: Beyond efficient transformer for long sequence time-series forecasting,” in Proceedings of the 35th AAAI Conference on Artificial Intelligence (AAAI) , vol. 35, no. 12, Virtual Conference, 2021, pp...

  44. [52]

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

    Y . Nie, N. H. Nguyen, P. Sinthong, and J. Kalagnanam, “A time series is worth 64 words: Long-term forecasting with transformers,” in The Eleventh International Conference on Learning Representations , 2023

Pith tools

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