pith. sign in

arxiv: 2605.21550 · v1 · pith:QXLIM3C4new · submitted 2026-05-20 · 💻 cs.LG

PeakFocus: Bridging Peak Localization and Intensity Regression via a Unified Multi-Scale Framework for Electricity Load Forecasting

Wangzhi Yu , Peng Zhu , Qing Zhao , Yiwen Jiang , Dawei Cheng This is my paper

Pith reviewed 2026-05-22 00:24 UTC · model grok-4.3

classification 💻 cs.LG
keywords electricity load forecastingpeak localizationintensity regressionmulti-scale featuresunified frameworktime series predictiongrid scheduling
0
0 comments X

The pith

PeakFocus unifies peak timing localization and intensity regression for electricity load forecasting using a multi-scale framework and location-aware decoding.

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

Electricity load peak forecasting requires simultaneous prediction of when peaks occur and how intense they are. Current approaches treat these as separate stages, which breaks the connection between timing and strength estimates and often produces misjudged peaks or overly smooth intensity values. The paper introduces PeakFocus, which employs a single pipeline supervised by a triple hybrid loss, mixes coarse and fine features through a cascade to correct timing errors, and feeds detected timing information directly into the intensity decoder. Experiments on both a public electricity dataset and a large industrial load dataset show gains in both timing precision and intensity accuracy over existing baselines. If the unification holds, joint supervision of the two subtasks can replace the predict-then-locate pattern in peak-related time-series tasks.

Core claim

PeakFocus establishes a unified peak-aware pipeline that applies a triple hybrid loss to supervise temporal localization and intensity regression together. A Multi-Scale Mixing Peak Locator uses coarse features to reduce local-fluctuation misjudgments and cascades them into fine-grained features to fix timing misalignment. A Location-Aware Decoder then injects the resulting peak timing context into the intensity regression branch to counteract global smoothing and raise peak intensity accuracy.

What carries the argument

Unified Peak-Aware Pipeline that jointly optimizes localization and regression via triple hybrid loss, Multi-Scale Mixing Peak Locator for coarse-to-fine feature injection, and Location-Aware Decoder that supplies timing context to intensity estimation.

If this is right

  • Grid operators can schedule reserves with tighter timing windows because peak occurrence and magnitude are estimated together.
  • The tolerance-based evaluation protocol gives a practical success metric that tolerates small timing offsets rather than requiring exact matches.
  • Multi-scale cascade injection can be reused in other time-series models where local noise masks rare high-value events.
  • Explicit timing context supplied to regression reduces the dominance of average trends that otherwise flatten extreme values.

Where Pith is reading between the lines

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

  • Similar joint localization-regression supervision could be tested on other peaky signals such as traffic surges or financial volatility spikes.
  • The cascade mixing mechanism suggests that explicit coarse-to-fine pathways may help any multi-resolution time-series architecture that currently trains separate heads.
  • If timing context proves useful for intensity, the reverse direction—using intensity estimates to refine localization—could be added as a further consistency constraint.

Load-bearing premise

Jointly supervising temporal localization and intensity regression with a triple hybrid loss plus multi-scale feature injection will fix peak misjudgment and intensity smoothing without creating new trade-offs or dataset-specific biases.

What would settle it

An ablation that removes the Location-Aware Decoder and measures whether peak intensity mean absolute error rises measurably on the WLEL dataset.

Figures

Figures reproduced from arXiv: 2605.21550 by Dawei Cheng, Peng Zhu, Qing Zhao, Wangzhi Yu, Yiwen Jiang.

Figure 1
Figure 1. Figure 1: An illustration of limitations in PatchTST for ELPF [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The proposed PeakFocus architecture. An encoder extracts input features. MSM-PL resolves localization conflicts [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Efficiency radar on WLEL (H=336). Five axes compare quality (F1, MSE, BCS) and efficiency (#Params, inference latency). PeakFocus leads all quality axes while remaining competitive on both parameter count and inference latency. Efficiency Analysis. As shown in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Ablation study on WLEL and ELC (H=336). Five variants: PeakFocus (full); w/o MSM-PL (linear layer replacing pyramid); w/o LAD (no peak timing injection); w/o UPAP (standard MSE only); Hard mask (binary mask replacing soft Gaussian). Metrics: F1 ↑, BCS↓, TP-MSE↓; values averaged over 5 seeds. and substantially improves both timing precision and balanced accuracy; K=3 yields only marginal additional gains. T… view at source ↗
Figure 5
Figure 5. Figure 5: Cross-backbone generality of UPAP on WLEL and ELC ( [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Parameter sensitivity on WLEL (top) and ELC (bottom). (a) MSM-PL depth [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Internal mechanism visualization on WLEL ( [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison on the WLEL dataset ( [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
read the original abstract

Electricity load peak forecasting (ELPF), simultaneously predicting peak timing and intensity, is a prerequisite for effective grid scheduling and risk management. However, existing methods face three limitations. First, they adopt a two-stage predict-then-locate paradigm, which severs the link between temporal localization and intensity regression. Second, they still struggle with the multi-scale representation conflict, leading to peak misjudgment and timing misalignment. Third, the lack of explicit peak timing context during intensity regression causes intensity smoothing because predictions are dominated by global smoothing trends. To address these limitations, we propose PeakFocus, a unified framework for ELPF. (i) A Unified Peak-Aware Pipeline (UPAP) utilizes a triple hybrid loss to jointly supervise temporal localization and intensity regression, alongside a tolerance-based evaluation protocol. (ii) A Multi-Scale Mixing Peak Locator (MSM-PL) exploits coarse-grained features to mitigate peak misjudgment caused by local fluctuations, and injects them into fine-grained features via a cascade mechanism to resolve timing misalignment. (iii) A Location-Aware Decoder (LAD) injects peak timing context into the intensity regression process, providing explicit guidance to counteract intensity smoothing and improve peak intensity estimation. Extensive experiments on the public Electricity (ELC) dataset and our industrial-scale World Large-scale Electricity Load (WLEL) dataset show that PeakFocus outperforms baselines in both timing precision and intensity estimation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces PeakFocus, a unified multi-scale framework for electricity load peak forecasting (ELPF) that simultaneously addresses peak timing localization and intensity regression. It proposes (i) a Unified Peak-Aware Pipeline (UPAP) employing a triple hybrid loss for joint supervision of localization and regression together with a tolerance-based evaluation protocol, (ii) a Multi-Scale Mixing Peak Locator (MSM-PL) that cascades coarse-grained features into fine-grained ones to reduce peak misjudgment and timing misalignment, and (iii) a Location-Aware Decoder (LAD) that injects explicit peak timing context into intensity regression to counteract smoothing. The central empirical claim is that PeakFocus outperforms existing baselines on both the public Electricity (ELC) dataset and the industrial-scale World Large-scale Electricity Load (WLEL) dataset in timing precision and intensity estimation.

Significance. If the reported gains are substantiated by rigorous ablations, error bars, and statistical tests, the work would be significant for grid scheduling and risk management. A unified treatment of localization and regression could reduce the disconnect inherent in two-stage pipelines and yield more reliable peak forecasts on both public and large-scale industrial data.

major comments (2)
  1. [§4 (Experiments)] §4 (Experiments) and associated tables: the central claim of joint improvement in timing precision and intensity estimation on both ELC and WLEL rests on the unverified premise that the triple hybrid loss, MSM-PL cascade, and LAD produce net gains rather than compensating errors or dataset-specific tuning. No ablation results isolating each component (e.g., performance when the cascade injection or location context is removed) are referenced, leaving the 'no new trade-offs' assumption unsupported.
  2. [§3.1 (UPAP)] §3.1 (UPAP): the tolerance-based evaluation protocol is introduced as part of the joint supervision but its precise definition, threshold selection, and interaction with the triple hybrid loss are not shown to be parameter-free or robust across the two datasets; this directly affects reproducibility of the reported outperformance.
minor comments (2)
  1. [Abstract] Abstract: while the three limitations and three proposed modules are clearly enumerated, the abstract contains no numerical results, error bars, or baseline names, which is atypical for an empirical claim of outperformance.
  2. [Notation and figures] Notation and figures: ensure all acronyms (UPAP, MSM-PL, LAD) are expanded on first use in the main text and that multi-scale cascade diagrams include explicit feature-dimension labels for clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to strengthen the empirical support and reproducibility.

read point-by-point responses

  1. Referee: [§4 (Experiments)] §4 (Experiments) and associated tables: the central claim of joint improvement in timing precision and intensity estimation on both ELC and WLEL rests on the unverified premise that the triple hybrid loss, MSM-PL cascade, and LAD produce net gains rather than compensating errors or dataset-specific tuning. No ablation results isolating each component (e.g., performance when the cascade injection or location context is removed) are referenced, leaving the 'no new trade-offs' assumption unsupported.

    Authors: We agree that explicit ablations are necessary to confirm net gains rather than compensating effects. In the revised version we will add a dedicated ablation subsection in §4 that isolates each component: (i) UPAP without MSM-PL cascade, (ii) full model without LAD location injection, and (iii) variants with/without the triple hybrid loss. Results will be reported on both ELC and WLEL with error bars and paired statistical tests. revision: yes

  2. Referee: [§3.1 (UPAP)] §3.1 (UPAP): the tolerance-based evaluation protocol is introduced as part of the joint supervision but its precise definition, threshold selection, and interaction with the triple hybrid loss are not shown to be parameter-free or robust across the two datasets; this directly affects reproducibility of the reported outperformance.

    Authors: We will expand §3.1 with the exact formulation of the tolerance-based protocol, including the mathematical definition of tolerated timing windows and how the tolerance interacts with each term of the triple hybrid loss. We will also report the threshold selection procedure (cross-validation on a held-out validation split) and include a sensitivity table demonstrating stable performance across a range of tolerance values on both datasets. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical model proposal validated on external datasets

full rationale

The paper proposes an architectural framework (UPAP with triple hybrid loss, MSM-PL cascade, LAD injection) and reports empirical gains on the public ELC dataset plus the authors' new WLEL dataset. No mathematical derivation chain, equations, or fitted parameters are shown that reduce the claimed timing/intensity improvements to the inputs by construction. The central claim rests on comparative experiments rather than self-definition, self-citation load-bearing, or renaming of known results. This is the normal case for an applied ML architecture paper; the derivation is self-contained against the reported benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The framework implicitly assumes that multi-scale features and timing context can be injected without distorting the underlying load signal.

pith-pipeline@v0.9.0 · 5796 in / 1078 out tokens · 29945 ms · 2026-05-22T00:24:48.168148+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages · 1 internal anchor

  1. [1]

    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,” inAAAI, vol. 35, 2021, pp. 11 106–11 115

    work page 2021

  2. [2]

    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 NeurIPS, vol. 34, 2021, pp. 22 419–22 430

    work page 2021

  3. [3]

    TimeMixer: Decomposable multiscale mixing for time series forecasting,

    S. Wang, H. Wu, X. Shi, T. Hu, H. Luo, L. Ma, J. Y . Zhang, and J. Zhou, “TimeMixer: Decomposable multiscale mixing for time series forecasting,” inICLR, 2024

    work page 2024

  4. [4]

    Adaptive multi-scale decomposition framework for time series forecasting,

    Y . Hu, P. Liu, P. Zhu, D. Cheng, and T. Dai, “Adaptive multi-scale decomposition framework for time series forecasting,” inAAAI, vol. 39, 2025, pp. 17 359–17 367

    work page 2025

  5. [5]

    Deep Time Series Models: A Comprehensive Survey and Benchmark

    Y . Wang, H. Wu, J. Dong, Y . Liu, C. Wang, M. Long, and J. Wang, “Deep time series models: A comprehensive survey and benchmark,” arXiv preprint arXiv:2407.13278, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  6. [6]

    Deep learning for time series forecasting: Tutorial and literature survey,

    K. Benidis, S. S. Rangapuram, V . Flunkert, Y . Wang, D. Maddix, C. Turkmen, J. Gasthaus, M. Bohlke-Schneider, D. Salinas, L. Stella, F.-X. Aubet, L. Callot, and T. Januschowski, “Deep learning for time series forecasting: Tutorial and literature survey,”ACM Comput. Surv., vol. 55, no. 6, pp. 1–36, 2022

    work page 2022

  7. [7]

    Daily peak electrical load forecasting with a multi-resolution approach,

    Y . Amara-Ouali, M. Fasiolo, Y . Goude, and H. Yan, “Daily peak electrical load forecasting with a multi-resolution approach,”Int. J. Forecast., vol. 39, no. 3, pp. 1272–1286, 2023

    work page 2023

  8. [8]

    MetaEformer: Unveiling and leveraging meta-patterns for complex and dynamic systems load forecasting,

    S. Huang, T. Zhang, Z. Zhang, X. Wang, L. Wang, and X. Wang, “MetaEformer: Unveiling and leveraging meta-patterns for complex and dynamic systems load forecasting,” inKDD, 2025, pp. 991–1002

    work page 2025

  9. [9]

    Electrical peak demand forecasting - A review,

    S. Dai, F. Meng, H. Dai, Q. Wang, and X. Chen, “Electrical peak demand forecasting - A review,”arXiv preprint arXiv:2108.01393, 2021

    work page arXiv 2021

  10. [10]

    Unlocking the potential of deep learning in peak-hour series forecasting,

    Z. Zhang, X. Wang, J. Xie, H. Zhang, and Y . Gu, “Unlocking the potential of deep learning in peak-hour series forecasting,” inCIKM, 2023, pp. 4415–4419

    work page 2023

  11. [11]

    Enhancing wind power forecasting at local peak points: A novel Seq2LPP model,

    N. Zhu, Y . Wang, K. Yuan, Y . Pan, and K. Zhang, “Enhancing wind power forecasting at local peak points: A novel Seq2LPP model,”IEEE Trans. Ind. Informat., 2025, early Access

    work page 2025

  12. [12]

    Reversible instance normalization for accurate time-series forecasting against distribution shift,

    T. Kim, J. Kim, Y . Tae, C. Park, J.-H. Choi, and J. Choo, “Reversible instance normalization for accurate time-series forecasting against distribution shift,” inICLR, 2021

    work page 2021

  13. [13]

    Spatial-temporal identity: A simple yet effective baseline for multivariate time series forecasting,

    Z. Shao, Z. Zhang, F. Wang, and Y . Xu, “Spatial-temporal identity: A simple yet effective baseline for multivariate time series forecasting,” in CIKM, 2022, pp. 4455–4459

    work page 2022

  14. [14]

    On the role of attention masks and LayerNorm in Transformers,

    X. Wu, A. Ajorlou, Y . Wang, S. Jegelka, and A. Jadbabaie, “On the role of attention masks and LayerNorm in Transformers,” inNeurIPS, vol. 37, 2024, pp. 14 774–14 809

    work page 2024

  15. [15]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Doll´ar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” inCVPR, 2017, pp. 2117–2125

    work page 2017

  16. [16]

    U-Net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional networks for biomedical image segmentation,” inMICCAI, 2015, pp. 234–241

    work page 2015

  17. [17]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” inNeurIPS, vol. 30, 2017

    work page 2017

  18. [18]

    Focal loss for dense object detection,

    T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Doll ´ar, “Focal loss for dense object detection,” inICCV, 2017, pp. 2980–2988

    work page 2017

  19. [19]

    Density- based weighting for imbalanced regression,

    M. Steininger, K. Kobs, P. Davidson, A. Krause, and A. Hotho, “Density- based weighting for imbalanced regression,”Mach. Learn., vol. 110, pp. 2187–2211, 2021

    work page 2021

  20. [20]

    Learning stationary time series using Gaussian processes with nonparametric kernels,

    F. Tobar, T. D. Bui, and R. E. Turner, “Learning stationary time series using Gaussian processes with nonparametric kernels,” inNeurIPS, vol. 28, 2015

    work page 2015

  21. [21]

    Complex event recognition in the big data era: a survey,

    N. Giatrakos, E. Alevizos, A. Artikis, A. Deligiannakis, and M. N. Garofalakis, “Complex event recognition in the big data era: a survey,” VLDB J., vol. 29, no. 1, pp. 313–352, 2020

    work page 2020

  22. [22]

    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,” inICLR, 2023

    work page 2023

  23. [23]

    SegRNN: Segment recurrent neural network for long-term time series forecasting,

    S. Lin, W. Lin, W. Wu, F. Zhao, R. Mo, and H. Zhang, “SegRNN: Segment recurrent neural network for long-term time series forecasting,” IEEE IoT J., vol. 13, no. 5, pp. 9861–9871, 2026

    work page 2026

  24. [24]

    CycleNet: Enhancing time series forecasting through modeling periodic patterns,

    S. Lin, W. Lin, X. Hu, W. Wu, R. Mo, and H. Zhong, “CycleNet: Enhancing time series forecasting through modeling periodic patterns,” inNeurIPS, vol. 37, 2024, pp. 106 315–106 345

    work page 2024

  25. [25]

    Large language models for time series analysis: Methodologies, applications, and emerging challenges,

    W. Yu, D. Cheng, L. Zhu, and C. Jiang, “Large language models for time series analysis: Methodologies, applications, and emerging challenges,” TechRxiv, 2026

    work page 2026

  26. [26]

    Breaking information granularity heterogeneity: A mutual information-inspired causal discovery framework for multi-rate time series,

    K. Zhu, C. Zhao, and B. Huang, “Breaking information granularity heterogeneity: A mutual information-inspired causal discovery framework for multi-rate time series,”IEEE TKDE, 2025

    work page 2025

  27. [27]

    Unlocking the power of LSTM for long term time series forecasting,

    Y . Kong, Z. Wang, Y . Nie, T. Zhou, S. Zohren, Y . Liang, P. Sun, and Q. Wen, “Unlocking the power of LSTM for long term time series forecasting,” inAAAI, vol. 39, 2025, pp. 11 968–11 976

    work page 2025

  28. [28]

    Are Transformers effective for time series forecasting?

    A. Zeng, M. Chen, L. Zhang, and Q. Xu, “Are Transformers effective for time series forecasting?” inAAAI, vol. 37, 2023, pp. 11 121–11 128

    work page 2023

  29. [29]

    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,” inICLR, 2020

    work page 2020

  30. [30]

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

    T. Zhou, Z. Ma, Q. Wen, X. Wang, L. Sun, and R. Jin, “FEDformer: Frequency enhanced decomposed transformer for long-term series forecasting,” inICML, 2022, pp. 27 268–27 286

    work page 2022

  31. [31]

    iTrans- former: Inverted transformers are effective for time series forecasting,

    Y . Liu, T. Hu, H. Zhang, H. Wu, S. Wang, L. Ma, and M. Long, “iTrans- former: Inverted transformers are effective for time series forecasting,” inICLR, 2024

    work page 2024

  32. [32]

    TimesNet: Temporal 2D-variation modeling for general time series analysis,

    H. Wu, T. Hu, Y . Liu, H. Zhou, J. Wang, and M. Long, “TimesNet: Temporal 2D-variation modeling for general time series analysis,” in ICLR, 2023

    work page 2023

  33. [33]

    Pyraformer: Low-complexity pyramidal attention for long-range time series modeling and forecasting,

    S. Liu, H. Yu, C. Liao, J. Li, W. Lin, A. X. Liu, and S. Dustdar, “Pyraformer: Low-complexity pyramidal attention for long-range time series modeling and forecasting,” inICLR, 2022

    work page 2022

  34. [34]

    SCINet: Time series modeling and forecasting with sample convolution and interaction,

    M. Liu, A. Zeng, M. Chen, Z. Xu, Q. Lai, L. Ma, and Q. Xu, “SCINet: Time series modeling and forecasting with sample convolution and interaction,” inNeurIPS, vol. 35, 2022, pp. 5816–5828

    work page 2022

  35. [35]

    Multi-scale adaptive graph neural network for multivariate time series forecasting,

    L. Chen, D. Chen, Z. Shang, B. Wu, C. Zheng, B. Wen, and W. Zhang, “Multi-scale adaptive graph neural network for multivariate time series forecasting,”IEEE TKDE, vol. 35, no. 10, pp. 10 748–10 761, 2023

    work page 2023

  36. [36]

    Exploring progress in multivariate time series forecasting: Comprehensive benchmarking and heterogeneity analysis,

    Z. Shao, F. Wang, Y . Xu, W. Wei, C. Yu, Z. Zhang, D. Yao, T. Sun, G. Jin, X. Cao, G. Cong, C. S. Jensen, and X. Cheng, “Exploring progress in multivariate time series forecasting: Comprehensive benchmarking and heterogeneity analysis,”IEEE TKDE, vol. 37, no. 1, pp. 291–305, 2024

    work page 2024

  37. [37]

    Efficient multivariate time series forecasting via calibrated language models with privileged knowledge distillation,

    C. Liu, H. Miao, Q. Xu, S. Zhou, C. Long, Y . Zhao, Z. Li, and R. Zhao, “Efficient multivariate time series forecasting via calibrated language models with privileged knowledge distillation,” inICDE, 2025, pp. 3165– 3178

    work page 2025

  38. [38]

    Towards spatio- temporal aware traffic time series forecasting,

    R.-G. Cirstea, B. Yang, C. Guo, T. Kieu, and S. Pan, “Towards spatio- temporal aware traffic time series forecasting,” inICDE, 2022, pp. 2900– 2913

    work page 2022

  39. [39]

    A unified replay-based continuous learning framework for spatio-temporal prediction on streaming data,

    H. Miao, Y . Zhao, C. Guo, B. Yang, K. Zheng, F. Huang, J. Xie, and C. S. Jensen, “A unified replay-based continuous learning framework for spatio-temporal prediction on streaming data,” inICDE, 2024, pp. 1050–1062

    work page 2024

  40. [40]

    Multi-step spatio-temporal forecasting with decoupled dynamic graphs,

    K. Zhao, C. Guo, Y . Cheng, P. Han, M. Chen, and B. Yang, “Multi-step spatio-temporal forecasting with decoupled dynamic graphs,” inICDE, 2024, pp. 3142–3155

    work page 2024

  41. [41]

    Robust and explainable autoencoders for unsupervised time series outlier detection,

    T. Kieu, B. Yang, C. Guo, R.-G. Cirstea, Y . Zhao, Y . Song, and C. S. Jensen, “Robust and explainable autoencoders for unsupervised time series outlier detection,” inICDE, 2022, pp. 1342–1354

    work page 2022

  42. [42]

    TimeFilter: Patch-specific spatial-temporal graph filtration for time series forecasting,

    Y . Hu, G. Zhang, P. Liu, D. Lan, N. Li, D. Cheng, T. Dai, S.-T. Xia, and S. Pan, “TimeFilter: Patch-specific spatial-temporal graph filtration for time series forecasting,” inICML, 2025

    work page 2025

  43. [43]

    Meta-learning for cross-region electricity load forecasting under distribution shift,

    J. Hu, Y . Liu, C. Guo, B. Yang, and C. S. Jensen, “Meta-learning for cross-region electricity load forecasting under distribution shift,” inICDE, 2025, pp. 2018–2031

    work page 2025