pith. sign in

arxiv: 2606.22126 · v1 · pith:WR76BCGEnew · submitted 2026-06-20 · 💻 cs.CL

From Recognition to Understanding: Unlocking Cognitive Time Series Reasoning with LLMs

Xin Qiu , Junlong Tong , Yao Zhang , Yunpu Ma , Wei Zhang , Xiaoyu Shen This is my paper

Pith reviewed 2026-06-26 11:47 UTC · model grok-4.3

classification 💻 cs.CL
keywords time series reasoningLLM alignmentcognitive benchmarkpatch encodingmultimodal fusionTSCognitionTSAligntemporal decision making
0
0 comments X

The pith

TSAlign encodes time series as patches and aligns them to LLM semantic directions to support cognitive reasoning tasks.

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

Current LLM work on time series mostly reduces the problem to low-level curve fitting and prediction. The paper claims this misses the semantic, contextual, and reasoning demands of real decisions. It therefore builds TSCognition, a benchmark of 41K QA pairs spanning five tasks drawn from fifteen public sources, and introduces TSAlign to map patch representations of the series into the LLM embedding space through gated residual injection and multivariate fusion. Experiments show the resulting model beats LLM, VLM, and time-series QA baselines on both the new benchmark and TimerBed while using less compute. The central move is therefore to treat time series as objects that can be aligned with language rather than fitted as curves.

Core claim

Existing task formulations reduce time series understanding to curve-fitting systems focused on low-level prediction. TSCognition supplies a multimodal benchmark of roughly 41K QA samples across Decoding, Grounding, Inferring, Extrapolating, and Acting tasks collected from fifteen public sources. TSAlign encodes the series into compact patch-level representations and aligns them with semantic directions in the LLM embedding space via gated residual injection and multivariate fusion, yielding higher accuracy than prior LLM, VLM, and time-series QA methods on TSCognition and TimerBed at substantially lower computational cost.

What carries the argument

TSAlign framework that encodes time series into compact patch-level representations and aligns them with semantic directions in the LLM embedding space via gated residual injection and multivariate fusion.

If this is right

  • LLMs become usable for tasks that combine time series with textual context such as inferring causes or selecting actions.
  • Patch-level encoding plus gated alignment reduces the compute needed for multimodal time-series reasoning compared with vision-language models.
  • Performance gains on TimerBed indicate the alignment approach transfers beyond the newly constructed benchmark.
  • The five-task structure supplies a concrete testbed for measuring progress from recognition to understanding in temporal data.

Where Pith is reading between the lines

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

  • The same patch-alignment recipe could be tested on other sequential modalities such as audio or video event streams.
  • If the alignment holds, downstream systems could generate natural-language explanations of time-series anomalies without separate post-hoc modules.
  • The benchmark construction method suggests a template for turning existing public datasets into reasoning QA collections in other scientific domains.

Load-bearing premise

The five cognitive tasks and 41K QA samples drawn from fifteen public sources accurately represent the semantic, contextual, and reasoning demands of real-world temporal decision-making.

What would settle it

A new collection of real-world time-series decision scenarios whose ground-truth answers require contextual inference or action planning, on which TSAlign shows no accuracy gain over standard curve-fitting or direct LLM prompting baselines.

Figures

Figures reproduced from arXiv: 2606.22126 by Junlong Tong, Wei Zhang, Xiaoyu Shen, Xin Qiu, Yao Zhang, Yunpu Ma.

Figure 1
Figure 1. Figure 1: Limitations of existing time series QA datasets (upper) vs. our TSCognition (bottom). Our [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Using a cloud system workload scenario as an example, we construct five hierarchical tasks [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of TSAlign. TSAlign encodes multivariate time series into patch-level represen [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Efficiency analysis. (a) Token composition (TS token vs. Text token); (b) Average inference [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: NSPE and RFD analysis of modality alignment. The left and middle plots show NSPE for [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: QA construction pipeline of TSCognition dynamics, reason with semantic context, and support task-oriented judgment. Compared with traditional tasks that primarily focus on forecasting error or classification accuracy, this setting encourages the development of more interpretable time series intelligence systems that are closer to real-world analytical needs. TSAlign has broad potential applications, includ… view at source ↗
Figure 7
Figure 7. Figure 7: Distribution of TSCognition samples across eight domains and five hierarchical tasks. [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Prompt template for TS-Text baseline. Prompt Template for Vision-Text Baseline Instruction: You are an expert in time series analysis and visual reasoning. Please answer the multiple-choice question based on the time series plot, context, and candidate answers. Time Series Image: [Line plot of the time series] Variable Information: [Variable names and descriptions] Context: [Scenario background or task con… view at source ↗
Figure 9
Figure 9. Figure 9: Prompt template for Vision-Text baseline. [PITH_FULL_IMAGE:figures/full_fig_p026_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Prompt template for TSAlign. Formally, let E ∈ R |V|×d denote the word embedding matrix of the pretrained LLM, where |V| is the vocabulary size and d is the embedding dimension. We perform PCA on E and take the top-k principal components to form an orthonormal basis Uk ∈ R d×k . Given a time series representation z ∈ R d , its projection onto the dominant language subspace is given by UkU⊤ k z, and the re… view at source ↗
Figure 11
Figure 11. Figure 11: Example of the Decoding task. 32 [PITH_FULL_IMAGE:figures/full_fig_p032_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Example of the Grounding task. 33 [PITH_FULL_IMAGE:figures/full_fig_p033_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Example of the Inferring task. 34 [PITH_FULL_IMAGE:figures/full_fig_p034_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Example of the Extrapolating task. 35 [PITH_FULL_IMAGE:figures/full_fig_p035_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Example of the Acting task. 36 [PITH_FULL_IMAGE:figures/full_fig_p036_15.png] view at source ↗
read the original abstract

Time series analysis has recently been coupled with Large Language Models (LLMs) to leverage their reasoning and world knowledge capabilities, yet gains remain limited. We attribute this to a fundamental mismatch between existing task formulations and LLM strengths: most settings reduce time series understanding to curve-fitting systems, focusing on low-level prediction while ignoring the semantic, contextual, and reasoning-intensive nature of real-world temporal decision-making.To address these limitations, we introduce TSCognition, a multimodal benchmark for multi-dimensional time series reasoning. It collects real-world time series and textual information from 15 public sources and constructs approximately 41K QA samples around five cognitive reasoning tasks: Decoding, Grounding, Inferring, Extrapolating, and Acting. Building on this, we further propose TSAlign, a unified framework that encodes time series into compact patch-level representations and aligns them with semantic directions in the LLM embedding space via gated residual injection and multivariate fusion.Experiments show that TSAlign outperforms existing LLM, VLM, and time series QA baselines on TSCognition and the publicly available TimerBed benchmark while substantially reducing computational cost.Code is available at: [https://github.com/EIT-NLP/CognitiveTSR](https://github.com/EIT-NLP/CognitiveTSR)

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 / 1 minor

Summary. The manuscript introduces TSCognition, a multimodal benchmark for cognitive time series reasoning constructed from 15 public sources yielding ~41K QA samples across five tasks (Decoding, Grounding, Inferring, Extrapolating, Acting). It proposes TSAlign, a framework that encodes time series into compact patch-level representations and aligns them to LLM embedding space via gated residual injection and multivariate fusion. The central claim is that TSAlign outperforms existing LLM, VLM, and time series QA baselines on TSCognition and TimerBed while reducing computational cost.

Significance. If the empirical claims hold with proper controls and the benchmark tasks genuinely capture semantic/contextual reasoning demands beyond curve-fitting, the work could meaningfully redirect LLM-based time series research toward higher-level temporal decision-making. The reuse of public data sources and provision of code support reproducibility and are positive features.

major comments (2)
  1. [Abstract] Abstract: The statement that TSAlign 'outperforms existing LLM, VLM, and time series QA baselines on TSCognition and the publicly available TimerBed benchmark while substantially reducing computational cost' supplies no numerical metrics, baseline specifications, ablation results, or statistical tests. This absence is load-bearing for the central empirical claim and prevents assessment of effect sizes or reliability.
  2. [Abstract] Abstract: The five cognitive reasoning tasks and 41K QA samples are motivated as addressing a mismatch with real-world temporal decision-making, yet no details are given on task construction, inter-annotator validation, or explicit differentiation from standard time series QA formulations. This leaves the benchmark's fidelity to the stated motivation unverified and central to interpreting any performance gains.
minor comments (1)
  1. The GitHub repository link is provided, which is helpful for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the two major comments on the abstract below and will make corresponding revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The statement that TSAlign 'outperforms existing LLM, VLM, and time series QA baselines on TSCognition and the publicly available TimerBed benchmark while substantially reducing computational cost' supplies no numerical metrics, baseline specifications, ablation results, or statistical tests. This absence is load-bearing for the central empirical claim and prevents assessment of effect sizes or reliability.

    Authors: We agree that the abstract would be strengthened by including concrete metrics. In the revised version we will add specific performance numbers (accuracy/F1 gains on TSCognition and TimerBed), name the primary baselines, and reference the ablation studies plus statistical tests reported in Sections 4–5. revision: yes

  2. Referee: [Abstract] Abstract: The five cognitive reasoning tasks and 41K QA samples are motivated as addressing a mismatch with real-world temporal decision-making, yet no details are given on task construction, inter-annotator validation, or explicit differentiation from standard time series QA formulations. This leaves the benchmark's fidelity to the stated motivation unverified and central to interpreting any performance gains.

    Authors: The abstract is intentionally concise; the full construction details, data sources, and differentiation from standard QA appear in Section 3. We will expand the abstract with a brief description of the five tasks, the 15 public sources, and the cognitive focus. Inter-annotator validation is not applicable to our largely automated construction pipeline, but we can add a sentence clarifying this approach. revision: partial

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper's central claims rest on constructing TSCognition (41K QA samples across five tasks from 15 public sources) and the TSAlign framework (patch-level encoding with gated residual injection and multivariate fusion), followed by empirical comparisons against LLM/VLM/time-series baselines on TSCognition and TimerBed. No equations, derivations, or self-citations are shown that reduce any prediction or result to a quantity defined by the authors' own prior work or fitted inputs by construction. The contribution is self-contained via new benchmark creation and standard benchmarking, with no load-bearing steps that collapse to self-definition or renaming.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review performed on abstract only; no explicit free parameters, axioms, or invented entities are stated. The central claim rests on the unverified premise that the five named tasks capture the intended cognitive gap.

axioms (1)
  • domain assumption Existing time-series-plus-LLM tasks reduce understanding to curve-fitting and therefore fail to engage LLM reasoning strengths.
    Stated in the opening motivation; used to justify the new benchmark design.

pith-pipeline@v0.9.1-grok · 5758 in / 1374 out tokens · 24633 ms · 2026-06-26T11:47:11.217200+00:00 · methodology

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

52 extracted references · 9 linked inside Pith

  1. [1]

    Brown, B

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, et al. Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901, 2020

    1901

  2. [2]

    J. Chen, A. Feng, Z. Zhao, J. Garza, G. Nurbek, C. Qin, A. Maatouk, L. Tassiulas, Y . Gao, and R. Ying. Mtbench: A multimodal time series benchmark for temporal reasoning and question answering.arXiv preprint arXiv:2503.16858, 2025

    arXiv 2025

  3. [3]

    Cheng, J

    M. Cheng, J. Wang, D. Wang, X. Tao, Q. Liu, and E. Chen. Can slow-thinking llms reason over time? empirical studies in time series forecasting. InProceedings of the Nineteenth ACM International Conference on Web Search and Data Mining, pages 99–110, 2026

    2026

  4. [4]

    D. C. Dowson and B. Landau. The fréchet distance between multivariate normal distributions. Journal of multivariate analysis, 12(3):450–455, 1982

    1982

  5. [5]

    Ethayarajh

    K. Ethayarajh. How contextual are contextualized word representations? comparing the geometry of bert, elmo, and gpt-2 embeddings. InProceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), pages 55–65, 2019

    2019

  6. [6]

    T. Guan, Z. Meng, D. Li, S. Wang, C.-H. H. Yang, Q. Wen, Z. Liu, S. M. Siniscalchi, M. Jin, and S. Pan. Timeomni-1: Incentivizing complex reasoning with time series in large language models. arXiv preprint arXiv:2509.24803, 2025

    Pith/arXiv arXiv 2025

  7. [7]

    T. Guan, S. Pan, J. Barthelemy, Z. Li, Y . Cai, C. Alippi, M. Jin, and S. Pan. Timeomni-vl: Unified models for time series understanding and generation.arXiv preprint arXiv:2602.17149, 2026

    Pith/arXiv arXiv 2026

  8. [8]

    Z. He, S. Alnegheimish, and M. Reimherr. Harnessing vision-language models for time series anomaly detection. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 21690–21698, 2026

    2026

  9. [9]

    Heusel, H

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium.Advances in neural information processing systems, 30, 2017

    2017

  10. [10]

    S. Jia, B. Song, C. Ye, and C. Yuan. M3time: Llm-enhanced multi-modal, multi-scale, and multi-frequency multivariate time series forecasting. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 22265–22273, 2026

    2026

  11. [11]

    Jiang, Y

    Y . Jiang, Y . Chen, X. Li, Q. Chao, S. Liu, and G. Cong. Fstllm: Spatio-temporal llm for few shot time series forecasting. InForty-second International Conference on Machine Learning, 2025

    2025

  12. [12]

    Jiang, K

    Y . Jiang, K. Ning, Z. Pan, X. Shen, J. Ni, W. Yu, A. Schneider, H. Chen, Y . Nevmyvaka, and D. Song. Multi-modal time series analysis: A tutorial and survey. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 2, pages 6043–6053, 2025

    2025

  13. [13]

    M. Jin, S. Wang, L. Ma, Z. Chu, J. Zhang, X. Shi, P.-Y . Chen, Y . Liang, Y .-F. Li, S. Pan, and Q. Wen. Time-llm: Time series forecasting by reprogramming large language models. In B. Kim, Y . Yue, S. Chaudhuri, K. Fragkiadaki, M. Khan, and Y . Sun, editors,International Conference on Learning Representations, volume 2024, pages 23857–23880, 2024

    2024

  14. [14]

    B. Jing, S. Chen, L. Zheng, B. Liu, Z. Li, J. Zou, T. Wei, Z. Liu, Z. Zeng, R. Qiu, et al. Tsaqa: Time series analysis question and answering benchmark.arXiv preprint arXiv:2601.23204, 2026

    Pith/arXiv arXiv 2026

  15. [15]

    Y . Kong, Y . Yang, Y . Hwang, W. Du, S. Zohren, Z. Wang, M. Jin, and Q. Wen. Time-mqa: Time series multi-task question answering with context enhancement. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 29736–29753, 2025

    2025

  16. [16]

    Z. Li, S. Li, and X. Yan. Time series as images: Vision transformer for irregularly sampled time series.Advances in Neural Information Processing Systems, 36:49187–49204, 2023. 11

    2023

  17. [17]

    C. Liu, Q. Xu, H. Miao, S. Yang, L. Zhang, C. Long, Z. Li, and R. Zhao. Timecma: Towards llm-empowered multivariate time series forecasting via cross-modality alignment. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 18780–18788, 2025

    2025

  18. [18]

    H. Liu, C. Liu, and B. A. Prakash. A picture is worth a thousand numbers: Enabling llms reason about time series via visualization. InProceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 7486–7518, 2025

    2025

  19. [19]

    H. Liu, S. Xu, Z. Zhao, L. Kong, H. Kamarthi, A. B. Sasanur, M. Sharma, J. Cui, Q. Wen, C. Zhang, et al. Time-mmd: Multi-domain multimodal dataset for time series analysis.Advances in Neural Information Processing Systems, 37:77888–77933, 2024

    2024

  20. [20]

    P. Liu, H. Guo, T. Dai, N. Li, J. Bao, X. Ren, Y . Jiang, and S.-T. Xia. Calf: Aligning llms for time series forecasting via cross-modal fine-tuning. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 18915–18923, 2025

    2025

  21. [21]

    W. Liu, H. Wu, X. Qiu, Y . Fan, Y . Zhang, A. Zhao, Y . Ma, and X. Shen. Vica: Efficient multimodal llms with vision-only cross-attention.arXiv preprint arXiv:2602.07574, 2026

    Pith/arXiv arXiv 2026

  22. [22]

    Y . Liu, T. Hu, H. Zhang, H. Wu, S. Wang, L. Ma, and M. Long. itransformer: Inverted transformers are effective for time series forecasting.arXiv preprint arXiv:2310.06625, 2023

    Pith/arXiv arXiv 2023

  23. [23]

    Y . Luo, Y . Zhou, M. Cheng, J. Wang, D. Wang, T. Pan, and J. Zhang. Time series forecasting as reasoning: A slow-thinking approach with reinforced llms.arXiv preprint arXiv:2506.10630, 2025

    Pith/arXiv arXiv 2025

  24. [24]

    Meunier, F

    R. Meunier, F. Benamara, V . Moriceau, Z. Qiao, and S. Ramasamy. Crisists: Coupling social media textual data and meteorological time series for urgency classification. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 16082–16099, 2025

    2025

  25. [25]

    Mu and P

    J. Mu and P. Viswanath. All-but-the-top: Simple and effective postprocessing for word repre- sentations. InInternational Conference on Learning Representations, 2018

    2018

  26. [26]

    Nagrath and S

    S. Nagrath and S. K. Panigrahy. Patch-level tokenization with cnn encoders and attention for improved transformer time-series forecasting.arXiv preprint arXiv:2601.12467, 2026

    arXiv 2026

  27. [27]

    J. Ni, Z. Zhao, C. Shen, H. Tong, D. Song, W. Cheng, D. Luo, and H. Chen. Harnessing vision models for time series analysis: a survey. InProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, pages 10612–10620, 2025

    2025

  28. [28]

    Y . Nie, N. H. Nguyen, P. Sinthong, and J. Kalagnanam. A time series is worth 64 words: Long-term forecasting with transformers.arXiv preprint arXiv:2211.14730, 2022

    Pith/arXiv arXiv 2022

  29. [29]

    Z. Pan, Y . Jiang, S. Garg, A. Schneider, Y . Nevmyvaka, and D. Song.S2ip-llm: Semantic space informed prompt learning with llm for time series forecasting. InForty-first International Conference on Machine Learning, 2024

    2024

  30. [30]

    Prithyani, M

    V . Prithyani, M. Mohammed, R. Gadgil, R. Buitrago, V . Jain, and A. Chadha. On the feasibility of vision-language models for time-series classification.arXiv preprint arXiv:2412.17304, 2024

    arXiv 2024

  31. [31]

    Z. Qiao, S. Pan, A. Wang, V . Zhukova, Y . Liu, X. Jiang, Q. Wen, M. Long, M. Jin, and C. Liu. It’s time: Towards the next generation of time series forecasting benchmarks.arXiv preprint arXiv:2602.12147, 2026

    Pith/arXiv arXiv 2026

  32. [32]

    X. Qiu, J. Tong, Y . Sun, Y . Ma, and X. Shen. The few govern the many: Unveiling few-layer dominance for time series models.arXiv preprint arXiv:2511.07237, 2025

    arXiv 2025

  33. [33]

    X. Qiu, J. Tong, Y . Sun, Y . Ma, W. Zhang, and X. Shen. Rethinking the role of llms in time series forecasting.arXiv preprint arXiv:2602.14744, 2026. 12

    arXiv 2026

  34. [34]

    Qwen, :, A. Yang, B. Yang, B. Zhang, B. Hui, B. Zheng, B. Yu, C. Li, D. Liu, F. Huang, H. Wei, H. Lin, J. Yang, J. Tu, J. Zhang, J. Yang, J. Yang, J. Zhou, J. Lin, K. Dang, K. Lu, K. Bao, K. Yang, L. Yu, M. Li, M. Xue, P. Zhang, Q. Zhu, R. Men, R. Lin, T. Li, T. Tang, T. Xia, X. Ren, X. Ren, Y . Fan, Y . Su, Y . Zhang, Y . Wan, Y . Liu, Z. Cui, Z. Zhang, ...

    2025

  35. [35]

    Schumacher, E

    D. Schumacher, E. Nourbakhsh, R. Slavin, and A. Rios. Prompting underestimates llm capability for time series classification.arXiv preprint arXiv:2601.03464, 2026

    arXiv 2026

  36. [36]

    M. Tan, M. A. Merrill, V . Gupta, T. Althoff, and T. Hartvigsen. Are language models actually useful for time series forecasting?Advances in Neural Information Processing Systems, 37:60162– 60191, 2024

    2024

  37. [37]

    Tarasiou, E

    M. Tarasiou, E. Chavez, and S. Zafeiriou. Vits for sits: Vision transformers for satellite image time series. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10418–10428, 2023

    2023

  38. [38]

    J. Tong, L. Xie, S. Fang, W. Yang, and K. Zhang. Hourly solar irradiance forecasting based on encoder–decoder model using series decomposition and dynamic error compensation.Energy Conversion and Management, 270:116049, 2022

    2022

  39. [39]

    Y . Wang, P. Lei, J. Song, Y . Hao, T. Chen, Y . Zhang, L. Jia, Y . Li, and Z. Wei. Itformer: Bridging time series and natural language for multi-modal qa with large-scale multitask dataset. In International Conference on Machine Learning, pages 63324–63344. PMLR, 2025

    2025

  40. [40]

    H. Wu, T. Hu, Y . Liu, H. Zhou, J. Wang, and M. Long. Timesnet: Temporal 2d-variation modeling for general time series analysis. InThe Eleventh International Conference on Learning Representations, 2022

    2022

  41. [41]

    W. Wu, Z. Zhang, L. Liu, X. Xu, J. Zhuang, K. Fan, Q. Lv, J. Liu, C. Zhang, Z. Yuan, et al. Scits: Scientific time series understanding and generation with llms.arXiv preprint arXiv:2510.03255, 2025

    arXiv 2025

  42. [42]

    X. Wu, J. Lu, Z. Li, X. Qiu, J. Hu, C. Guo, C. S. Jensen, and B. Yang. Timeart: Towards agentic time series reasoning via tool-augmentation.arXiv preprint arXiv:2601.13653, 2026

    arXiv 2026

  43. [43]

    Z. Xie, Z. Li, X. He, L. Xu, X. Wen, T. Zhang, J. Chen, R. Shi, and D. Pei. Chatts: Aligning time series with llms via synthetic data for enhanced understanding and reasoning.Proceedings of the VLDB Endowment, 18(8):2385–2398, 2025

    2025

  44. [44]

    G. Xu, P. Jin, Z. Wu, H. Li, Y . Song, L. Sun, and L. Yuan. Llava-cot: Let vision language models reason step-by-step. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 2087–2098, 2025

    2087

  45. [45]

    X. Xu, H. Wang, Y . Liang, P. S. Yu, Y . Zhao, and K. Shu. Can multimodal llms perform time series anomaly detection? InProceedings of the ACM Web Conference 2026, pages 5392–5403, 2026

    2026

  46. [46]

    F. Yu, X. Guo, L. Yuan, H. Kang, H. Zhao, L. Qin, F. Huang, B. Hu, and T. Zhou. Tsrbench: A comprehensive multi-task multi-modal time series reasoning benchmark for generalist models.arXiv preprint arXiv:2601.18744, 2026

    Pith/arXiv arXiv 2026

  47. [47]

    Z. Yue, Y . Wang, J. Duan, T. Yang, C. Huang, Y . Tong, and B. Xu. Ts2vec: Towards universal representation of time series. InProceedings of the AAAI conference on artificial intelligence, volume 36, pages 8980–8987, 2022

    2022

  48. [48]

    Zhang, Q

    S. Zhang, Q. Fang, Z. Yang, and Y . Feng. Llava-mini: Efficient image and video large multimodal models with one vision token. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  49. [49]

    L. N. Zheng, W. Liang, W. E. Zhang, M. Xu, O. Maennel, and W. Chen. Lifting manifolds to mitigate pseudo-alignment in llm4ts. InProceedings of the ACM Web Conference 2026, pages 3764–3775, 2026. 13

    2026

  50. [50]

    Zhong, W

    S. Zhong, W. Ruan, M. Jin, H. Li, Q. Wen, and Y . Liang. Time-vlm: Exploring multimodal vision-language models for augmented time series forecasting. InInternational Conference on Machine Learning, pages 78478–78497. PMLR, 2025

    2025

  51. [51]

    T. Zhou, Z. Ma, Q. Wen, X. Wang, L. Sun, and R. Jin. Fedformer: Frequency enhanced decomposed transformer for long-term series forecasting. InInternational conference on machine learning, pages 27268–27286. PMLR, 2022

    2022

  52. [52]

    perceiving signals

    T. Zhou, P. Niu, L. Sun, R. Jin, et al. One fits all: Power general time series analysis by pretrained lm.Advances in neural information processing systems, 36:43322–43355, 2023. A Ethics Statement Our study is limited to methodological and empirical investigation and does not involve human participants, animal subjects, or environmentally sensitive mater...

    2023