Pith. sign in

REVIEW 2 major objections 1 minor 86 references

TimeBlocks assembles lightweight time-series models from a reusable pool of modular blocks selected by routing and keeps a small representative subset for continual calibration.

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 · grok-4.3

2026-06-28 16:01 UTC pith:DVGJTLVU

load-bearing objection TimeBlocks puts forward a block pool plus routing and StreamCore subset method to build lightweight models for time-series streams. the 2 major comments →

arxiv 2606.02142 v1 pith:DVGJTLVU submitted 2026-06-01 cs.LG cs.DB

TimeBlocks: Foundational and Continual Time-Series Blockbase -- Extended Version

David Campos , Bin Yang , Tung Kieu , Lei Chen , Chenjuan Guo , Christian S. Jensen This is my paper
classification cs.LG cs.DB
keywords time-seriesmodular modelsfoundational modelscontinual learningdata streamsmodel routinglightweight modelsstream approximation
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.

The paper aims to create time-series models that handle multiple tasks like foundational models but remain small enough for real-time stream processing and repeated updates. It keeps a collection of interchangeable model blocks and uses a routing process to combine them into task-specific models when new data arrives. StreamCore extracts a compact subset from the incoming stream that approximates the full data well enough to support ongoing recalibration. If successful, this would produce accurate models that adapt without storing everything or using massive fixed architectures. The approach targets settings where large models fail because of size and lack of support for continuous adjustment.

Core claim

TimeBlocks enables versatile time-series processing by maintaining a pool of interchangeable modular model blocks that a routing strategy iteratively selects to construct lightweight accurate models, equipped with StreamCore to build a representative small subset preserving a guaranteed approximation of the stream for continual calibration, outperforming baselines on multiple datasets and tasks.

What carries the argument

A pool of interchangeable modular model blocks selected iteratively by a routing strategy, together with StreamCore for constructing a representative stream subset.

Load-bearing premise

A routing strategy can reliably pick blocks that yield accurate models for any time-series data, and StreamCore's small subset continues to approximate the full stream well enough that performance does not degrade over time.

What would settle it

An experiment on new time-series streams where the routed block models fail to beat the baselines or where accuracy falls steadily as more data arrives despite repeated use of the StreamCore subset.

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

If this is right

  • Models become small enough for real-time responses under strict time and compute limits.
  • The same block pool supports multiple tasks by changing which blocks are chosen.
  • Continual calibration occurs using only the maintained subset instead of the entire history.
  • Models remain deployable in hardware-limited environments where large foundational models cannot run.
  • Performance exceeds standard baselines across the tested datasets and tasks.

Where Pith is reading between the lines

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

  • The block pool could be updated by adding or replacing individual blocks without rebuilding everything.
  • Similar routing over modular components might apply to other sequential data such as sensor readings or financial ticks.
  • The guaranteed approximation property of the subset could be checked periodically by comparing predictions on held-out recent data.
  • Edge devices could host the routing and subset logic locally while occasionally syncing block updates from a central pool.

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

2 major / 1 minor

Summary. The manuscript proposes TimeBlocks, a framework that maintains a pool of interchangeable modular model blocks from which a routing strategy iteratively assembles lightweight time-series models tailored to specific data and tasks. It augments this with StreamCore, which constructs a small representative subset of an incoming data stream that preserves a guaranteed approximation, enabling continual model calibration. Experiments across multiple datasets and tasks report that the resulting models outperform existing baselines.

Significance. If the routing mechanism reliably produces accurate models and StreamCore's subset construction maintains its approximation guarantee without performance degradation, the work could enable practical deployment of versatile time-series models in streaming, real-time, and hardware-constrained environments where large offline foundational models are unsuitable.

major comments (2)
  1. [Abstract] Abstract: the central claim that StreamCore 'preserves a guaranteed approximation of the stream over time' is load-bearing for the continual-calibration contribution, yet the abstract supplies neither the formal statement of the guarantee nor the construction algorithm; without these details the experimental outperformance cannot be assessed for robustness over long streams.
  2. [Abstract] Abstract: the routing strategy is described only at the level of 'iteratively selects the most suitable blocks'; because this selection process is the mechanism asserted to produce accurate lightweight models for arbitrary time-series data, the absence of selection criteria, objective function, or convergence argument leaves the reliability claim unsupported by the provided description.
minor comments (1)
  1. [Abstract] Abstract: the experimental study is summarized only as 'on multiple data sets and covering multiple tasks'; adding the number of datasets, tasks, and at least one quantitative performance delta would strengthen the claim of outperformance.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed comments on the abstract. Both points identify areas where the high-level summary can be strengthened without altering the manuscript's core claims. We address each below and will revise the abstract accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that StreamCore 'preserves a guaranteed approximation of the stream over time' is load-bearing for the continual-calibration contribution, yet the abstract supplies neither the formal statement of the guarantee nor the construction algorithm; without these details the experimental outperformance cannot be assessed for robustness over long streams.

    Authors: We agree the abstract should be more informative on this point. In revision we will add a single sentence stating the formal approximation guarantee (e.g., the subset maintains an ε-approximation in a chosen divergence or norm) and briefly name the construction procedure (e.g., the greedy coreset-style selection with periodic refresh). The full proof and algorithm remain in Section 4; the abstract change will not exceed the typical length limit. revision: yes

  2. Referee: [Abstract] Abstract: the routing strategy is described only at the level of 'iteratively selects the most suitable blocks'; because this selection process is the mechanism asserted to produce accurate lightweight models for arbitrary time-series data, the absence of selection criteria, objective function, or convergence argument leaves the reliability claim unsupported by the provided description.

    Authors: The abstract is intentionally concise, but we accept that a slightly more precise phrasing is warranted. We will revise the sentence to indicate that blocks are chosen by minimizing a task-specific loss (or validation error) over a small candidate pool at each iteration. The concrete objective, stopping criterion, and any convergence properties are already derived in Section 3; the abstract update will reference this mechanism at the same level of detail used for comparable routing methods in the literature. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper describes a modular block pool, routing strategy, and StreamCore subset construction as a methodological proposal whose performance is asserted via experimental results on external datasets and tasks. No equations, parameter-fitting procedures, or self-citations are presented that reduce any claimed prediction or guarantee to a tautological restatement of the inputs. The central claims rest on empirical outperformance rather than internal definitional closure or load-bearing self-reference.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only; no explicit free parameters, axioms, or invented entities are stated.

pith-pipeline@v0.9.1-grok · 5808 in / 1025 out tokens · 27048 ms · 2026-06-28T16:01:50.980083+00:00 · methodology

0 comments
read the original abstract

The ongoing digitization has led to a proliferation of time-series data streams that monitor a variety of processes, from which valuable insights may be obtained. Further, the emergence of successful foundational language models begs the question of whether it is possible to achieve time-series models with the foundational properties of handling multiple tasks, while being sufficiently lightweight to allow real-time data stream processing. Existing foundational time-series models are often large and only effective in offline settings without stringent time and computational constraints, and where repeated model calibration is not needed. However, when applied to data streams, these models are ineffective due to their size and lack of support for continual calibration, which compromise their ability to deliver accurate real-time responses, their durability, and their deployability in hardware-limited settings. We propose TimeBlocks to enable versatile time-series processing by facilitating the efficient building of lightweight models suitable for multiple tasks under variable conditions. In particular, the method maintains a pool of interchangeable and modular model blocks that can be used to construct new time-series models. When presented with specific time-series data, a routing strategy iteratively selects the most suitable blocks to construct a lightweight and accurate model for the data. We equip TimeBlocks with a method called StreamCore to build a representative small subset of the data stream, which preserves a guaranteed approximation of the stream over time, enabling continual model calibration. An experimental study on multiple data sets and covering multiple tasks shows that TimeBlocks enables to build models capable of outperforming existing baselines.

Figures

Figures reproduced from arXiv: 2606.02142 by Bin Yang, Chenjuan Guo, Christian S. Jensen, David Campos, Lei Chen, Tung Kieu.

Figure 1
Figure 1. Figure 1: Paradigms for Time-Series Processing. (a) In the specialized model paradigm, a new model must be trained for each [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: TimeBlocks Paradigm Overview. 3.2 Blockbase Routing 3.2.1 Overview. While numerous time-series models may be avail￾able for supporting multiple tasks, as shown in [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Multi-Scale Block Processing. In addition to enable a better processing of a time series, this multi-scale strategy ensures that the blocks included in the Block￾base are capable of managing time series with multiple scales. Output Standardization: In order to achieve independence and interchangeability among every block, it is important to establish a common interface that allows the interaction and conne… view at source ↗
Figure 5
Figure 5. Figure 5: Router Overview. 𝑅𝑜𝑢𝑡𝑒𝑟(𝑟𝑝 ) ≈ 𝑏 1 𝑝+1 . This router aims to approximate the fingerprint of a block that matches the output 𝑟𝑝 . Furthermore, each block is designed to produce a standardized output (see Section 3.2.2), ensuring that the mapping remains consistent across all residuals. Block Selection Problem: The router is a heuristic proposed to efficiently identify a set of blocks, a process that is othe… view at source ↗
Figure 6
Figure 6. Figure 6: Router Training and Block Selection. 3.2.5 Inference-Time Model Building. To efficiently build an infer￾ence-time model, we use the Blockbase in conjunction with the router to select the blocks that are most suitable for processing a given time series. The router first selects the optimal cluster for processing the time series, thereby reducing the number of blocks to be considered for model construction. … view at source ↗
Figure 8
Figure 8. Figure 8: (a) when 𝐽 = 2. Model Size (MB) Autoformer 2449.604 PatchTST 1920.776 TimeMixer 25.476 TimesNet 241.680 Lag-Llama 576.528 Moment 151.641 AnomalyTran 29.472 TTM 3.176 TimeBlocks 𝐽 = 6 2.985 TimeBlocks 𝐽 = 2 0.995 (a) All Baselines. 1 2 3 0.27 0.28 0.29 0.3 𝐽 = 1 𝐽 = 2 𝐽 = 3 𝐽 = 4 𝐽 = 5 𝐽 = 6 𝐽 = 6 Size (MB) MSE TimeBlocks TTM (b) Number of Blocks, ETTh2 [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Model Efficiency. Forecasting horizon is 96. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 7
Figure 7. Figure 7: Classification Ranking (Accuracy). 4.2.7 Model Size. When comparing the average model size among data sets, see [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 10
Figure 10. Figure 10: Model Efficiency. Forecasting horizon is 96. [PITH_FULL_IMAGE:figures/full_fig_p009_10.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

86 extracted references · 4 canonical work pages · 1 internal anchor

  1. [1]

    Maddix, Hao Wang, Michael W

    Abdul Fatir Ansari, Lorenzo Stella, Ali Caner Türkmen, Xiyuan Zhang, Pedro Mercado, Huibin Shen, Oleksandr Shchur, Syama Sundar Rangapuram, Sebastian Pineda-Arango, Shubham Kapoor, Jasper Zschiegner, Danielle C. Maddix, Hao Wang, Michael W. Mahoney, Kari Torkkola, Andrew Gordon Wilson, Michael Bohlke-Schneider, and Bernie Wang. 2024. Chronos: Learning the...

    2024

  2. [2]

    Arjun Ashok, Étienne Marcotte, Valentina Zantedeschi, Nicolas Chapados, and Alexandre Drouin. 2024. TACTiS-2: Better, Faster, Simpler Attentional Copulas for Multivariate Time Series. InICLR

    2024

  3. [3]

    Ran Avnimelech and Nathan Intrator. 1999. Boosted Mixture Of Experts: An Ensemble Learning Scheme.Neural Comput.11, 2 (1999), 483–497

    1999

  4. [4]

    Vladimir Braverman, Dan Feldman, Harry Lang, Daniela Rus, and Adiel Statman

  5. [5]

    Least-Mean-Squares Coresets for Infinite Streams.IEEE Trans. Knowl. Data Eng.35, 9 (2023), 8699–8712

    2023

  6. [6]

    Pietro Buzzega, Matteo Boschini, Angelo Porrello, Davide Abati, and Simone Calderara. 2020. Dark Experience for General Continual Learning: a Strong, Simple Baseline. InNeurIPS

    2020

  7. [7]

    David Campos, Bin Yang, Tung Kieu, Miao Zhang, Chenjuan Guo, and Chris- tian S. Jensen. 2024. QCore: Data-Efficient, On-Device Continual Calibration for Quantized Models.Proc. VLDB Endow.17, 11 (2024), 2708–2721

    2024

  8. [8]

    David Campos, Miao Zhang, Bin Yang, Tung Kieu, Chenjuan Guo, and Christian S. Jensen. 2023. LightTS: Lightweight Time Series Classification with Adaptive Ensemble Distillation.Proc. ACM Manag. Data1, 2 (2023), 171:1–171:27

    2023

  9. [9]

    Rich Caruana, Alexandru Niculescu-Mizil, Geoff Crew, and Alex Ksikes. 2004. Ensemble selection from libraries of models. InICML, Vol. 69

    2004

  10. [10]

    Olivares, Boris N

    Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza Ramírez, Max Mergenthaler Canseco, and Artur Dubrawski. 2023. NHITS: Neural Hierar- chical Interpolation for Time Series Forecasting. InAAAI. 6989–6997

    2023

  11. [11]

    Arslan Chaudhry, Marc’Aurelio Ranzato, Marcus Rohrbach, and Mohamed Elho- seiny. 2019. Efficient Lifelong Learning with A-GEM. InICLR

    2019

  12. [12]

    Schneider, Eduard H

    Sang Keun Choe, Hwijeen Ahn, Juhan Bae, Kewen Zhao, Minsoo Kang, Youngseog Chung, Adithya Pratapa, Willie Neiswanger, Emma Strubell, Teruko Mitamura, Jeff G. Schneider, Eduard H. Hovy, Roger B. Grosse, and Eric P. Xing

  13. [13]

    What is Your Data Worth to GPT? LLM-Scale Data Valuation with Influence Functions.CoRRabs/2405.13954 (2024)

    work page arXiv 2024

  14. [14]

    Ben Cohen, Emaad Khwaja, Kan Wang, Charles Masson, Elise Ramé, Youssef Dou- bli, and Othmane Abou-Amal. 2024. Toto: Time Series Optimized Transformer for Observability.CoRRabs/2407.07874 (2024)

    work page arXiv 2024

  15. [15]

    Vincent Cohen-Addad, Niklas Hjuler, Nikos Parotsidis, David Saulpic, and Chris Schwiegelshohn. 2019. Fully Dynamic Consistent Facility Location. InNeurIPS. 3250–3260. David Campos et al

    2019

  16. [16]

    Abhimanyu Das, Weihao Kong, Rajat Sen, and Yichen Zhou. 2024. A Decoder- only Foundation Model for Time-series Forecasting. InICML, Vol. 235. 10148– 10167

    2024

  17. [17]

    Bagnall, Kaveh Kamgar, Chin-Chia Michael Yeh, Yan Zhu, Shaghayegh Gharghabi, Chotirat Ann Ratanamahatana, and Eamonn J

    Hoang Anh Dau, Anthony J. Bagnall, Kaveh Kamgar, Chin-Chia Michael Yeh, Yan Zhu, Shaghayegh Gharghabi, Chotirat Ann Ratanamahatana, and Eamonn J. Keogh. 2019. The UCR time series archive.IEEE CAA J. Autom. Sinica6, 6 (2019), 1293–1305

    2019

  18. [18]

    Vijay Ekambaram, Arindam Jati, Nam Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam. 2023. TSMixer: Lightweight MLP-Mixer Model for Multivariate Time Series Forecasting. InKDD. 459–469

    2023

  19. [19]

    Nguyen, Pankaj Dayama, Chandra Reddy, Wesley M

    Vijay Ekambaram, Arindam Jati, Nam H. Nguyen, Pankaj Dayama, Chandra Reddy, Wesley M. Gifford, and Jayant Kalagnanam. 2024. Tiny Time Mixers (TTMs): Fast Pre-trained Models for Enhanced Zero/Few-Shot Forecasting of Multivariate Time Series. InNeurIPS

    2024

  20. [20]

    Schmidt, Jonathan Weber, Geoffrey I

    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 Min. Knowl. Discov.34, 6 (2020), 1936–1962

    2020

  21. [21]

    Dan Feldman and Michael Langberg. 2011. A unified framework for approximat- ing and clustering data. InSTOC. 569–578

    2011

  22. [22]

    Dan Feldman and Tamir Tassa. 2015. More Constraints, Smaller Coresets: Con- strained Matrix Approximation of Sparse Big Data. InSIGKDD. 249–258

    2015

  23. [23]

    Milton Friedman. 1940. A Comparison of Alternative Tests of Significance for the Problem of $m$ Rankings.Annals of Mathematical Statistics11 (1940), 86–92

    1940

  24. [24]

    Challu, Laurent Callot, Lenon Minorics, and Andrey Kan

    Mononito Goswami, Cristian I. Challu, Laurent Callot, Lenon Minorics, and Andrey Kan. 2023. Unsupervised Model Selection for Time Series Anomaly Detection. InICLR

    2023

  25. [25]

    Mononito Goswami, Konrad Szafer, Arjun Choudhry, Yifu Cai, Shuo Li, and Artur Dubrawski. 2024. MOMENT: A Family of Open Time-series Foundation Models. InICML

    2024

  26. [26]

    Nate Gruver, Marc Finzi, Shikai Qiu, and Andrew Gordon Wilson. 2023. Large Language Models Are Zero-Shot Time Series Forecasters. InNeurIPS

    2023

  27. [27]

    Jensen, and Bin Yang

    Chenjuan Guo, Christian S. Jensen, and Bin Yang. 2014. Towards Total Traffic Awareness.SIGMOD Rec.43, 3 (2014), 18–23

    2014

  28. [28]

    Sture Holm. 1979. A Simple Sequentially Rejective Multiple Test Procedure. Scandinavian Journal of Statistics6, 2 (1979), 65–70

    1979

  29. [29]

    J. A. Hoogeveen. 1991. Analysis of Christofides’ heuristic: Some paths are more difficult than cycles.Oper. Res. Lett.10, 5 (1991), 291–295

    1991

  30. [30]

    Kyle Hundman, Valentino Constantinou, Christopher Laporte, Ian Colwell, and Tom Söderström. 2018. Detecting Spacecraft Anomalies Using LSTMs and Non- parametric Dynamic Thresholding. InSIGKDD. 387–395

    2018

  31. [31]

    2010.K-Means Clustering

    Xin Jin and Jiawei Han. 2010.K-Means Clustering. Springer US, Boston, MA, 563–564

    2010

  32. [32]

    Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2021. Billion-Scale Similarity Search with GPUs.IEEE Trans. Big Data7, 3 (2021), 535–547

    2021

  33. [33]

    Andrej Karpathy. 2019. Multi-task learning in the wilderness. InICML. https: //slideslive.com/38917690/multitask-learning-in-the-wilderness

    work page arXiv 2019

  34. [34]

    Chi, Jeffrey Dean, and Neoklis Polyzotis

    Tim Kraska, Alex Beutel, Ed H. Chi, Jeffrey Dean, and Neoklis Polyzotis. 2018. The Case for Learned Index Structures. InSIGMOD. 489–504

    2018

  35. [35]

    Max Dupré la Tour, Monika Henzinger, and David Saulpic. 2024. Fully Dynamic k-Means Coreset in Near-Optimal Update Time. InESA (LIPIcs, Vol. 308). 100:1– 100:16

    2024

  36. [36]

    Yuening Li, Zhengzhang Chen, Daochen Zha, Mengnan Du, Jingchao Ni, Denghui Zhang, Haifeng Chen, and Xia Hu. 2022. Towards Learning Disentangled Repre- sentations for Time Series. InSIGKDD. 3270–3278

    2022

  37. [37]

    Yiming Li, Yanyan Shen, and Lei Chen. 2022. Camel: Managing Data for Efficient Stream Learning. InSIGMOD. 1271–1285

    2022

  38. [38]

    Jensen, and Bin Yang

    Zhe Li, Xiangfei Qiu, Peng Chen, Yihang Wang, Hanyin Cheng, Yang Shu, Jilin Hu, Chenjuan Guo, Aoying Zhou, Christian S. Jensen, and Bin Yang. 2025. TSFM- Bench: A Comprehensive and Unified Benchmark of Foundation Models for Time Series Forecasting. InKDD. 5595–5606

    2025

  39. [39]

    Jiaming Liang, Lei Cao, Samuel Madden, Zack Ives, and Guoliang Li. 2024. RITA: Group Attention is All You Need for Timeseries Analytics.Proc. ACM Manag. Data2, 1 (2024), 62:1–62:28

    2024

  40. [40]

    Zhiyu Liang, Jianfeng Zhang, Chen Liang, Hongzhi Wang, Zheng Liang, and Lujia Pan. 2023. A Shapelet-based Framework for Unsupervised Multivariate Time Series Representation Learning.Proc. VLDB Endow.17, 3 (2023), 386–399

    2023

  41. [41]

    Ilya Loshchilov and Frank Hutter. 2019. Decoupled Weight Decay Regularization. InICLR

    2019

  42. [42]

    Hua Lu, Bin Yang, and Christian S. Jensen. 2011. Spatio-temporal joins on symbolic indoor tracking data. InICDE. 816–827

    2011

  43. [43]

    Alaa Maalouf, Gilad Eini, Ben Mussay, Dan Feldman, and Margarita Osadchy

  44. [44]

    Neural Networks Learn

    A Unified Approach to Coreset Learning.IEEE Trans. Neural Networks Learn. Syst.35, 5 (2024), 6893–6905

    2024

  45. [45]

    Mathur and Nils Ole Tippenhauer

    Aditya P. Mathur and Nils Ole Tippenhauer. 2016. SWaT: a water treatment testbed for research and training on ICS security. InCySWater@CPSWeek. 31–36

    2016

  46. [46]

    Nimrod Megiddo and Kenneth J. Supowit. 1984. On the Complexity of Some Common Geometric Location Problems.SIAM J. Comput.13, 1 (1984), 182–196

    1984

  47. [47]

    Adam Meyerson. 2001. Online Facility Location. InFOCS. 426–431

    2001

  48. [48]

    Bilmes, and Jure Leskovec

    Baharan Mirzasoleiman, Jeff A. Bilmes, and Jure Leskovec. 2020. Coresets for Data-efficient Training of Machine Learning Models. InICML, Vol. 119. 6950– 6960

    2020

  49. [49]

    Guanlin Mo, Shihong Song, and Hu Ding. 2024. Towards Metric DBSCAN: Exact, Approximate, and Streaming Algorithms.Proc. ACM Manag. Data2, 3 (2024), 178

    2024

  50. [50]

    Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam

    Yuqi Nie, Nam H. Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam. 2023. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers. In ICLR

    2023

  51. [51]

    Oreshkin, Dmitri Carpov, Nicolas Chapados, and Yoshua Bengio

    Boris N. Oreshkin, Dmitri Carpov, Nicolas Chapados, and Yoshua Bengio. 2020. N-BEATS: Neural Basis Expansion Analysis for Interpretable Time Series Fore- casting. InICLR

    2020

  52. [52]

    Phillips

    Jeff M. Phillips. 2017. Coresets and Sketches. InHandbook of Discrete and Computational Geometry, Third Edition, Csaba D. Toth, Joseph O’Rourke, and Jacob E. Goodman (Eds.). Chapman and Hall/CRC, New York, 1269–1288

    2017

  53. [53]

    Lennart Oswald Purucker, Lennart Schneider, Marie Anastacio, Joeran Beel, Bernd Bischl, and Holger H. Hoos. 2023. Q(D)O-ES: Population-based Quality (Diversity) Optimisation for Post Hoc Ensemble Selection in AutoML. InAutoML (Proceedings of Machine Learning Research, Vol. 224). 10/1–34

    2023

  54. [54]

    Jensen, Zhenli Sheng, and Bin Yang

    Xiangfei Qiu, Jilin Hu, Lekui Zhou, Xingjian Wu, Junyang Du, Buang Zhang, Chenjuan Guo, Aoying Zhou, Christian S. Jensen, Zhenli Sheng, and Bin Yang

  55. [55]

    VLDB Endow.17, 9 (2024), 2363–2377

    TFB: Towards Comprehensive and Fair Benchmarking of Time Series Forecasting Methods.Proc. VLDB Endow.17, 9 (2024), 2363–2377

    2024

  56. [56]

    Kashif Rasul, Arjun Ashok, Andrew Robert Williams, Arian Khorasani, George Adamopoulos, Rishika Bhagwatkar, Marin Bilos, Hena Ghonia, Nadhir Vincent Hassen, Anderson Schneider, Sahil Garg, Alexandre Drouin, Nicolas Chapados, Yuriy Nevmyvaka, and Irina Rish. 2023. Lag-Llama: Towards Foundation Models for Time Series Forecasting. InNeurIPS

    2023

  57. [57]

    Lei Rui, Xiangdong Huang, Shaoxu Song, Yuyuan Kang, Chen Wang, and Jianmin Wang. 2024. Time Series Representation for Visualization in Apache IoTDB.Proc. ACM Manag. Data2, 1 (2024), 35:1–35:26

    2024

  58. [58]

    Schapire

    Robert E. Schapire. 2003.The Boosting Approach to Machine Learning: An Overview. Springer, New York, 149–171

    2003

  59. [59]

    Ya Su, Youjian Zhao, Chenhao Niu, Rong Liu, Wei Sun, and Dan Pei. 2019. Robust Anomaly Detection for Multivariate Time Series through Stochastic Recurrent Neural Network. InSIGKDD. 2828–2837

    2019

  60. [60]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurélien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lam- ple. 2023. LLaMA: Open and Efficient Foundation Language Models.CoRR abs/2302.13971 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  61. [61]

    Murad Tukan, Cenk Baykal, Dan Feldman, and Daniela Rus. 2021. On coresets for support vector machines.Theor. Comput. Sci.890 (2021), 171–191

    2021

  62. [62]

    Zhang, and Jun Zhou

    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. InICLR

    2024

  63. [63]

    Wenxuan Wang, Kai Wu, Yujian Betterest Li, Dan Wang, and Xiaoyu Zhang. 2025. Synthetic Series-Symbol Data Generation for Time Series Foundation Models. In NeurIPS

    2025

  64. [64]

    Yihang Wang, Yuying Qiu, Peng Chen, Yang Shu, Zhongwen Rao, Lujia Pan, Bin Yang, and Chenjuan Guo. 2025. LightGTS: A Lightweight General Time Series Forecasting Model. InICML

    2025

  65. [65]

    Yihang Wang, Yuying Qiu, Peng Chen, Kai Zhao, Yang Shu, Zhongwen Rao, Lujia Pan, Bin Yang, and Chenjuan Guo. 2025. Towards a General Time Series Forecasting Model with Unified Representation and Adaptive Transfer. InICML

    2025

  66. [66]

    Zhiguang Wang, Weizhong Yan, and Tim Oates. 2017. Time series classification from scratch with deep neural networks: A strong baseline. InIJCNN. 1578–1585

    2017

  67. [67]

    Frank Wilcoxon. 1945. Individual Comparisons by Ranking Methods.Biometrics Bulletin1, 6 (1945), 80–83

    1945

  68. [68]

    Gerald Woo, Chenghao Liu, Akshat Kumar, Caiming Xiong, Silvio Savarese, and Doyen Sahoo. 2024. Unified Training of Universal Time Series Forecasting Transformers. InICML, Vol. 235. 53140–53164

    2024

  69. [69]

    Gerald Woo, Chenghao Liu, Akshat Kumar, Caiming Xiong, Silvio Savarese, and Doyen Sahoo. 2024. Unified Training of Universal Time Series Forecasting Transformers. InICML (Proceedings of Machine Learning Research). 53140–53164

    2024

  70. [70]

    Haixu Wu, Tengge Hu, Yong Liu, Hang Zhou, Jianmin Wang, and Mingsheng Long. 2023. TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis. InICLR

    2023

  71. [71]

    Haixu Wu, Jiehui Xu, Jianmin Wang, and Mingsheng Long. 2021. Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Fore- casting. InNeurIPS. 22419–22430

    2021

  72. [72]

    Xingjian Wu, Jianxin Jin, Wanghui Qiu, Peng Chen, Yang Shu, Bin Yang, and Chenjuan Guo. 2026. Aurora: Towards Universal Generative Multimodal Time Series Forecasting. InICLR

    2026

  73. [73]

    Xingjian Wu, Xiangfei Qiu, Hongfan Gao, Jilin Hu, Bin Yang, and Chenjuan Guo. 2025. K2VAE: A Koopman-Kalman Enhanced Variational AutoEncoder for Probabilistic Time Series Forecasting. InICML. TimeBlocks: Foundational and Continual Time-Series Blockbase—Extended Version

    2025

  74. [74]

    Jiehui Xu, Haixu Wu, Jianmin Wang, and Mingsheng Long. 2022. Anomaly Transformer: Time Series Anomaly Detection with Association Discrepancy. In ICLR

    2022

  75. [75]

    Sean Bin Yang, Jilin Hu, Chenjuan Guo, Bin Yang, and Christian S. Jensen. 2023. LightPath: Lightweight and Scalable Path Representation Learning. InKDD. 2999–3010

    2023

  76. [76]

    Yiyuan Yang, Chaoli Zhang, Tian Zhou, Qingsong Wen, and Liang Sun. 2023. DCdetector: Dual Attention Contrastive Representation Learning for Time Series Anomaly Detection. InKDD. 3033–3045

    2023

  77. [77]

    Zhihan Yue, Yujing Wang, Juanyong Duan, Tianmeng Yang, Congrui Huang, Yunhai Tong, and Bixiong Xu. 2022. TS2Vec: Towards Universal Representation of Time Series. InAAAI. 8980–8987

    2022

  78. [78]

    Ailing Zeng, Muxi Chen, Lei Zhang, and Qiang Xu. 2023. Are Transformers Effective for Time Series Forecasting?. InAAAI. 11121–11128

    2023

  79. [79]

    George Zerveas, Srideepika Jayaraman, Dhaval Patel, Anuradha Bhamidipaty, and Carsten Eickhoff. 2021. A Transformer-based Framework for Multivariate Time Series Representation Learning. InSIGKDD. 2114–2124

    2021

  80. [80]

    Gary Chan

    Shuhan Zhong, Sizhe Song, Weipeng Zhuo, Guanyao Li, Yang Liu, and S.-H. Gary Chan. 2024. A Multi-Scale Decomposition MLP-Mixer for Time Series Analysis. Proc. VLDB Endow.17, 7 (2024), 1723–1736

    2024

Showing first 80 references.