Pith. sign in

REVIEW 5 major objections 5 minor 3 cited by

ITFormer: Bridging Time Series and Natural Language for Multi-Modal QA with Large-Scale Multitask Dataset

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A small trained bridge can attach any time-series encoder to a frozen large language model for question answering.

desk verdict Useful new dataset, unproven architecture claims; needs fair baselines, text-only controls, and a resolution of the 110k vs 11k discrepancy before the results can be taken at face value. read the letter →

arxiv 2506.20093 v1 pith:MOWK6EVD submitted 2025-06-25 cs.CL

Yilin Wang , Peixuan Lei , Jie Song , Yuzhe Hao , Tao Chen , Yuxuan Zhang , Lei Jia , Yuanxiang Li
show 1 more author
Zhongyu Wei
This is my paper
classification cs.CL
keywords time-seriesquestionansweringmultimodallargelanguagemodelstemporal-textualalignmentEngineMT-QAITFormerfrozenmodelcross-modalfusionaero-enginehealthmonitoring
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

The paper tries to establish that natural-language question answering over raw time-series signals is solvable by a small trainable bridge between two frozen models. It defines the Time-Series QA task, releases EngineMT-QA, a multi-task benchmark of aero-engine sensor question-answer pairs, and proposes ITFormer, which keeps both the time-series encoder and the large language model frozen and trains only an alignment module (about 0.07% of the total parameters). ITFormer-7B is reported to win every metric on all four tasks — understanding, perception, reasoning, decision-making — beating multimodal APIs such as GPT-4o and Gemini as well as time-series-text and vision-text baselines. The paper also reports that pretraining on EngineMT-QA transfers to the unrelated TimeSeriesExam benchmark. The contribution, if it holds, is a recipe: freeze the big models, train the bridge, and any sensor stream gains a natural-language interface.

What carries the argument

The central object is the ITFormer alignment module, a lightweight trainable bridge composed of four named components: Time Token Position Encoding (hierarchical position information at temporal, channel, and segment levels), Learnable Instruct Tokens (query-derived tokens that carry task instructions into the fusion), Instruct Time Attention (a two-stage channel-then-time aggregation guided by those instructions), and Time Token as Language (injecting the fused tokens into the LLM's input sequence by replacing placeholder tokens). The module's job is to make the frozen LLM read time-series semantics as if they were words, so the only parameters updated during supervised fine-tuning are the bridge's own.

What would settle it

Run ITFormer on EngineMT-QA with the signal degraded while the question text stays identical — shuffle the sensor channels, permute the ten cycles, or replace the window with noise matched to its statistics. If accuracy remains high, the answers come from language priors rather than the signal. Two supporting checks settle the benchmark itself: an inter-annotator agreement study on a sample of pairs, and a direct count of released pairs to resolve the abstract's 110k versus the main text's 11k.

Watch

Extended reading notes

Core claim

On its own terms, the paper's claim is that temporal semantics can be written into the embedding space of a frozen LLM. ITFormer encodes multivariate sensor windows with a frozen PatchTST encoder, adds position encodings at three levels (time steps, channels, segments), and derives task-specific guidance from learnable instruct tokens that are prepended to the question and refined by self-attention. A two-stage Instruct Time Attention then aggregates channels for each time step and then aggregates time, both under that guidance, producing fused tokens that replace placeholder tokens in the question's embedding sequence so the frozen decoder treats them as language. With only the alignment module trained, ITFormer-7B achieves the best reported scores on every EngineMT-QA metric, including 88.69 Accuracy and F1 on reasoning and 58.04 Rouge-L on understanding, and the ablation study attributes the largest gains to the position encoding and the instruct attention.

Load-bearing premise

The load-bearing premise is that EngineMT-QA's ground-truth answers, drafted by language models and checked by experts, genuinely require reading the time-series signal; the paper reports no human baseline or agreement measure, so if the questions are answerable from templates or language priors, the reported gains would not show true cross-modal understanding.

Editorial extensions

If this is right

  • Any existing time-series encoder can gain a natural-language question-answering interface by training the small bridge, without retraining either the encoder or the LLM.
  • EngineMT-QA offers a reusable four-task benchmark, allowing future methods to be compared on temporal understanding, perception, reasoning, and decision-making over the same sensor data.
  • Results scale with the frozen LLM's size, so upgrading from a 0.5B to a 7B model improves temporal-textual accuracy at essentially no change to the trainable parameter budget.
  • Pretraining on EngineMT-QA raises accuracy on the domain-agnostic TimeSeriesExam, indicating the dataset is a transferable resource beyond aero engines.
  • The same framework handles open-ended generation and fixed-option classification, so one deployment covers both answer styles.

Reading between the lines

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

  • If the recipe generalizes, a single small adapter could add interactive natural-language analysis to medical, climate, and industrial sensing systems; the paper only demonstrates the pattern on engine data plus one domain-agnostic exam, so the breadth is an extrapolation.
  • The paper does not run any control that removes the signal while keeping the question fixed, so how much of the reported accuracy reflects genuine signal reading versus answerable patterns in the LLM-generated questions remains my open concern, not a claim the paper addresses.
  • The abstract states over 110k question-answer pairs while the dataset section states 11k; the actual released count needs verification before downstream work can rely on the benchmark's scale.
  • A natural testable extension is whether the same instruct-token, channel-then-time attention pattern transfers to other structured modalities, such as multi-sensor arrays, graphs, or time-frequency images.
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

5 major / 5 minor

Summary. The paper introduces the Time-Series Question Answering (Time-Series QA) task, a new benchmark dataset called EngineMT-QA built from the N-CMAPSS aero-engine dataset, and an alignment method called ITFormer that connects a frozen time-series encoder with a frozen LLM using only a small number of trainable parameters. The claimed contributions are threefold: a large-scale multi-task temporal-textual QA dataset, a lightweight cross-modal alignment architecture (with Time Token Position Encoding, Learnable Instruct Tokens, Instruct Time Attention, and Time Token as Language), and experimental results showing that ITFormer outperforms a range of adapted baselines on EngineMT-QA as well as on the external TimeSeriesExam benchmark.

Significance. If the results hold, the paper would provide a useful recipe for attaching time-series encoders to frozen LLMs for question answering and a new benchmark for temporal-textual reasoning. The lightweight design (claimed <1% trainable parameters), the public release of code and dataset, and the transfer experiment to TimeSeriesExam are positive aspects that support reproducibility and external validation. However, the significance is currently undercut by internal inconsistencies in the dataset scale, by an unfair comparison setup against API baselines, by missing control experiments that would validate the benchmark, and by dimensional errors in the core attention equations. These issues must be resolved before the claimed state-of-the-art results and the benchmark's validity can be accepted.

major comments (5)
  1. [Abstract / Section 1 vs. Section 5.1 / Figure 3] The paper states in the Abstract and Section 1 that EngineMT-QA contains 'over 110k question-answer pairs,' while Section 5.1 and Figure 3 both state that the dataset 'comprises 11k QA pairs.' This tenfold discrepancy is unresolved and directly affects the central 'large-scale multi-task dataset' contribution. The authors must correct the number and ensure consistency across the abstract, introduction, main text, and figures.
  2. [Section 5.2 / Table 1 / Appendix D] The comparison in Table 1 is not apples-to-apples: ITFormer is fine-tuned on the EngineMT-QA training split, while ChatGPT-4o and Gemini are evaluated zero-shot on plotted signal images (as described in Appendix D). The 'fair comparison' paragraph claims that all baselines were adapted and trained with the same steps and epochs, but API models cannot be fine-tuned in this setup. The large margins in Table 1 may therefore reflect the fine-tuning versus zero-shot gap rather than architectural superiority. Please either fine-tune or otherwise adapt the API baselines under the same training protocol, or add a zero-shot ITFormer baseline and explicitly discuss the difference.
  3. [Section 5.1 / Appendix A / Figure 3] The validity of EngineMT-QA as a measure of temporal-textual reasoning is unverified. For the multiple-choice Perception and Reasoning tasks, answers are derived from N-CMAPSS health labels, but the paper provides no text-only baseline (the question and options without the time series), no sensor-shuffled or time-permuted control, no human performance estimate, and no inter-annotator agreement for the LLM-generated and expert-validated questions. Without these controls, high accuracy could result from template artifacts or language priors rather than genuine temporal reasoning. The TimeSeriesExam transfer in Section 5.6 offers partial external support but does not validate the EngineMT-QA labels. Please add these controls and report them.
  4. [Section 4.1 / Eqs. (7)-(9)] The Channel Instruct Fusing equations are dimensionally inconsistent. With Q_channel in R^{n x d_k} and K_channel derived from H_T in R^{L' x V x d}, the product Q_channel K_channel^T in Eq. (8) gives an attention matrix indexed by (n, L'*V), but Eq. (9) indexes A_channel[q, v] with v ranging only over V and then sums over q and v to produce H_channel[l, k]. No reshaping or summation over L' is specified, so the equations do not define a valid channel aggregation. Please rewrite this step with explicit tensor shapes and operations.
  5. [Section 5.2 / Table 1] All performance claims in Table 1 appear to be based on a single train/test split with no error bars, no multiple seeds, and no statistical significance tests. Given that the headline result is that 'ITFormer-7B consistently achieves the best performance on every metric,' the authors should report means and standard deviations over at least 3-5 seeds (or bootstrap confidence intervals) to support the claimed differences, especially where margins are narrow (e.g., Perception accuracy 65.07 for ITFormer-7B versus 64.07 for ITFormer-3B).
minor comments (5)
  1. [Tables 1 and 2] The column headers use 'BLUE' in both tables; this should be 'BLEU'.
  2. [Abstract and Section 1] There are minor language errors, including 'a adaptable paradigm' in the Abstract and 'an innovate framework' in the contributions list; these should be corrected.
  3. [Appendix A] The dataset is referred to as 'N-MAPSS' in Appendix A but as 'N-CMAPSS' in the main text; please use the official dataset name consistently.
  4. [Appendix B / Figure 8] Figure 8 and its surrounding text contain unreadable encoding artifacts (e.g., '/uni00000038/uni00000051/...'), so the reported statistical indicators cannot be inspected. Please replace with a clean figure and caption.
  5. [References] The reference for MCAN-VQA, listed as 'Zhou Yu, J. Y., Yuhao Cui, D. T., and Tian, Q.', appears malformed; the author names should be corrected.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the central result is a held-out test-set comparison with external TimeSeriesExam transfer, so it is not forced by construction; residual concerns (self-built benchmark, 110k-vs-11k discrepancy) are validity issues rather than circularity.

full rationale

The paper's central claim is empirical: ITFormer-7B achieves the best accuracy/Rouge-L/BLEU on the EngineMT-QA test subset (Table 1, Section 5.2). The paper states that 'ITFormer models were trained on the EngineMT-QA dataset, with training conducted on the training subset and evaluation on the test subset,' so the headline numbers are held-out predictions, not fitted values renamed as predictions. No equation in Sections 3-4 defines the target metric in terms of a fitted parameter; the loss (Eq. 17) is cross-entropy over a training split and the reported metrics are computed on a separate test split. The transfer experiment on the external TimeSeriesExam benchmark (Section 5.6, Table 3) provides independent grounding: pre-training on EngineMT-QA improves Pattern accuracy (0.83 to 0.86) and Anomaly accuracy (0.84 to 0.89) over fine-tuning without it, and ITFormer beats external multimodal baselines such as GPT-4o on that benchmark. The self-citations (Wang et al. 2022, 2023, 2024c, 2025a, 2025b, several with overlapping authors) appear in the Introduction as domain context for PHM signal processing, not as load-bearing justification for the ITFormer architecture or its results; they are peer-reviewed journal papers, not an imported uniqueness theorem or ansatz authority. Residual concerns are validity issues, not circularity: EngineMT-QA is author-constructed (Appendix A) with LLM-generated questions validated by experts but without reported inter-annotator agreement or a text-only/signal-ablated control, so shortcut vulnerability is unmeasured; and the scale claim is internally inconsistent ('over 110k question-answer pairs' in the Abstract and Section 1 vs '11k QA pairs' in Section 5.1 and Figure 3), which should be resolved by the authors. Section 5.6's concluding sentence is also truncated ('These results suggest that ENGINEMT-QA captures transferable' with no completion), so the benchmark-validation claim is asserted without its closing support. None of these concerns constitute an exhibited reduction of a prediction to its inputs under the evidence bar, so the circularity score is low.

Assumptions & free parameters 4 free parameters · 3 assumptions · 1 invented entities

The central experimental claims rest on the assumption that EngineMT-QA labels are valid ground truth, on the assumption that a frozen LLM can consume injected continuous tokens, and on a handful of hyperparameters chosen by ablation. No new physical entities are proposed; the only novel construct, Learnable Instruct Tokens, is a trained parameter vector whose utility is evidenced only by the paper's own ablations.

free parameters (4)
  • LIT length n = 25
    Number of learnable instruct tokens, tuned in ablation (Section 5.4); performance is sensitive to this value.
  • ITFormer layer count = 2
    Number of transformer layers in the alignment module, tuned in ablation (Section 5.4).
  • Patch size / stride = 60
    PatchTST tokenization parameters set in Section 5.2.
  • SFT epochs = 2
    Supervised fine-tuning duration set in Section 5.2.
assumptions (3)
  • domain assumption N-CMAPSS sensor data and its fault/RUL labels are accurate enough to define ground truth for engine health QA.
    EngineMT-QA derives all labels and QA pairs from N-CMAPSS (Arias Chao et al., 2021), used in Appendix A.
  • domain assumption The LLM-generated and expert-validated QA pairs faithfully represent physical relationships between the time-series signals and the answers.
    Dataset construction uses DeepSeek/ChatGPT refinement plus expert inspection (Appendix A); no inter-annotator agreement or error rate is reported.
  • ad hoc to paper A frozen LLM can condition on continuous fused time-series tokens inserted into its embedding sequence via placeholder replacement.
    The Time Token as Language strategy (Eq. 12) assumes the LLM interprets the injected H_fusion as meaningful context; the paper provides no analysis of token utilization beyond end-task metrics.
invented entities (1)
  • Learnable Instruct Tokens (LIT)
    purpose: Task-conditioned query tokens prepended to the text representation to guide channel and time attention and later replace placeholders.
    These vectors exist only as trained parameters; the paper's ablations connect them to performance but provide no externally falsifiable prediction tied specifically to LIT.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ITFormer: Bridging Time Series and Natural Language for Multi-Modal QA with Large-Scale Multitask Dataset." pith.science (2026). https://pith.science/paper/MOWK6EVD

@misc{pith2026250620093,
  author       = {Pith},
  title        = {Pith review of: ITFormer: Bridging Time Series and Natural Language for Multi-Modal QA with Large-Scale Multitask Dataset},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MOWK6EVD}},
  note         = {Machine review of arXiv:2506.20093}
}
read the original abstract

Time-series data are critical in diverse applications, such as industrial monitoring, medical diagnostics, and climate research. However, effectively integrating these high-dimensional temporal signals with natural language for dynamic, interactive tasks remains a significant challenge. To address this, we introduce the Time-Series Question Answering (Time-Series QA) task and release EngineMT-QA, the first large-scale, multi-task, temporal-textual QA dataset designed to capture complex interactions between time-series signals and natural language. Building on this resource, we propose the Instruct Time Transformer (ITFormer), a novel framework that bridges time-series encoders with frozen large language models (LLMs). ITFormer effectively extracts, aligns, and fuses temporal and textual features, achieving a strong improvement in QA accuracy over strong baselines with fewer than 1\% additional trainable parameters. By combining computational efficiency with robust cross-modal modeling, our work establishes a adaptable paradigm for integrating temporal data with natural language, paving the way for new research and applications in multi-modal AI. More details about the project, including datasets and code, are available at: https://pandalin98.github.io/itformer_site/

Figures

Figures reproduced from arXiv: 2506.20093 by the authors.

Figure 1
Figure 1. EngineMT-QA: A large-scale QA dataset based on aero engine time-series signals. The proposed ITFormer framework seamlessly connects any time-series encoder with LLMs, modeling temporal-text interaction by embedding time-series semantics into natural language. A growing need exists for practical applications where users can interact with time-series data directly using natural lan￾guage. For example, a engineer might… view at source ↗
Figure 2
Figure 2. The design of ITFormer includes several key components: 1. Time Token Position Encoding (TPE), 2. Learnable Instruct Tokens (LIT), and the 3. Instruct Time Attention (ITA) mechanism. The framework also incorporates the 4. Time Token as Language (TAL) strategy, which represents temporal semantics as language tokens to enhance the expressiveness of fused representations. xi,t ∈ R V , the encoder produces a latent repr… view at source ↗
Figure 3
Figure 3. Overview of the EngineMT-QA dataset. This dataset comprises 11k QA pairs across four tasks: Understanding, Percep￾tion, Reasoning, and Decision-Making. Each task reflects specific operational requirements of engine maintenance, including sensor relationships, health state interpretation, trend prediction, and maintenance decision-making. time-series QA, delivering robust performance with minimal parameter tuning, an… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4 [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Impact of ITFormer layers and LIT length on model performance. A moderate number of ITFormer layers optimizes performance, while very short LIT lengths limit multimodal ex￾traction, and overly long lengths lead to sparse information. To evaluate the contribution of eac…
Figure 6
Figure 6. Figure 6: Computational Efficiency of ITFormer. (a) and (b) show inference speed across channel counts and sequence lengths, while (c) analyzes input question length’s impact on the LIT mechanism, highlighting efficiency gains [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Dataset Construction Flow:The multi-modal time-series dataset for aircraft engines is constructed from NASA’s N-MAPSS dataset, integrating raw flight data analysis, LLM-based refinement, and expert validation. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Statistical Indicators of the Dataset: Key statistical characteristics of the dataset, including distributions of sensor readings, fault occurrences, and engine health metrics. These indicators provide insights into data variability, trends, and coverage, ensuring a co…
Figure 9
Figure 9. Figure 9: Data Example: Example of sensor trend analysis and semantic feature extraction in the dataset. Temporal understanding captures variations in engine parameters (e.g., fuel flow, fan speed) with descriptors like ”rapid increase” and ”stable” to enhance interpretability. …
Figure 10
Figure 10. Figure 10: Input signal template: Example plots for three different cycles (a, b, c). 18 [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. TS-Haystack: A Multi-Task Retrieval Benchmark for Long-Context Time-Series Reasoning

    cs.LG 2026-02 unverdicted novelty 7.0 of 10

    TS-Haystack benchmark shows time-series language models degrade sharply on long contexts while an agentic retrieval system using classifier tools matches or beats them on 9 of 10 tasks.

  2. CLIR-Bench: Benchmarking Multimodal Question Answering over Irregular Clinical Time Series

    cs.CL 2026-07 conditional novelty 6.0 of 10

    CLIR-Bench shows generalist and time-series LLMs struggle to ground clinical answers in sparse irregular ICU evidence, with top accuracy near 50% and weak causal evidence use.

  3. Time Series Forecasting as Reasoning: A Slow-Thinking Approach with Reinforced LLMs

    cs.LG 2025-06 unverdicted novelty 6.0 of 10

    Time-R1 trains LLMs via supervised fine-tuning followed by reinforcement learning with a time-series-specific reward and non-uniform GRIP sampling to enable multi-step reasoning that improves forecasting accuracy.

Reference graph

Works this paper leans on

57 extracted references · 39 canonical work pages · cited by 3 Pith papers

  1. [1]

    L., and Parikh, D

    Antol, S., Agrawal, A., Lu, J., Mitchell, M., Batra, D., Zitnick, C. L., and Parikh, D. Vqa: Visual question answering. In Proceedings of the IEEE international conference on computer vision, pp.\ 2425--2433, 2015

    2015

  2. [2]

    Aircraft Engine Run-to-Failure Dataset under Real Flight Conditions for Prognostics and Diagnostics

    Arias Chao, M., Kulkarni, C., Goebel, K., and Fink, O. Aircraft Engine Run-to-Failure Dataset under Real Flight Conditions for Prognostics and Diagnostics . 6 0 (1): 0 5, 2021. ISSN 2306-5729

    work page 2021

  3. [3]

    Accurate medium-range global weather forecasting with 3d neural networks

    Bi, K., Xie, L., Zhang, H., Chen, X., Gu, X., and Tian, Q. Accurate medium-range global weather forecasting with 3d neural networks. Nature, 619 0 (7970): 0 533--538, 2023

    work page 2023

  4. [4]

    Timeseriesexam: A time series understanding exam for large language models

    Cai, Y., Zhang, C., Liu, Y., He, J., Wang, Z., Liu, Y., Long, G., Wang, Y., Lin, H., and Zhou, J. Timeseriesexam: A time series understanding exam for large language models. arXiv preprint arXiv:2410.14752, 2024. URL https://arxiv.org/abs/2410.14752

    arXiv 2024

  5. [5]

    Clements, M. P. and Mizon, G. E. Empirical analysis of macroeconomic time series: Var and structural models. European Economic Review, 35 0 (4): 0 887--917, 1991

    work page 1991

  6. [6]

    Instructblip: Towards general-purpose vision-language models with instruction tuning

    Dai, W., Li, J., Li, D., Tiong, A., Zhao, J., Wang, W., Li, B., Fung, P., and Hoi, S. Instructblip: Towards general-purpose vision-language models with instruction tuning. arxiv 2023. arXiv preprint arXiv:2305.06500, 2, 2023

    arXiv 2023

  7. [7]

    Z., Webb, G

    Darban, Z. Z., Webb, G. I., Pan, S., Aggarwal, C. C., and Salehi, M. CARLA : Self-supervised contrastive representation learning for time series anomaly detection. Pattern Recognition, 157: 0 110874, 2025

    work page 2025

  8. [8]

    M., Parikh, D., and Batra, D

    Das, A., Kottur, S., Gupta, K., Singh, A., Yadav, D., Moura, J. M., Parikh, D., and Batra, D. Visual dialog. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 326--335, 2017

    work page 2017

Show all 57 references
  1. [9]

    Dennis, D., Acar, D. A. E., Mandikal, V., Sadasivan, V. S., Saligrama, V., Simhadri, H. V., and Jain, P. Shallow rnn: accurate time-series classification on resource constrained devices. Advances in neural information processing systems, 32, 2019

    2019

  2. [10]

    Glm: General language model pretraining with autoregressive blank infilling and task-specific fine-tuning

    Du, Z., Qian, Y., Liu, X., Ding, M., Qiu, J., Yang, Z., and Tang, J. Glm: General language model pretraining with autoregressive blank infilling and task-specific fine-tuning. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (ACL), 2022

    2022

  3. [11]

    Units: A unified multi-task time series model

    Gao, S., Koker, T., Queen, O., Hartvigsen, T., Tsiligkaridis, T., and Zitnik, M. Units: A unified multi-task time series model. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

    2024

  4. [12]

    MOMENT : A family of open time-series foundation models

    Goswami, M., Szafer, K., Choudhry, A., Cai, Y., Li, S., and Dubrawski, A. MOMENT : A family of open time-series foundation models. In International Conference on Machine Learning, 2024

    2024

  5. [13]

    Vector quantization pretraining for eeg time series with random projection and phase alignment

    Gui, H., Li, X., and Chen, X. Vector quantization pretraining for eeg time series with random projection and phase alignment. In International Conference on Machine Learning, pp.\ 16731--16750. PMLR, 2024

    2024

  6. [14]

    Hamilton, J. D. Time series analysis. Princeton university press, 2020

    2020

  7. [15]

    Y., Wen, Q., Zambon, D., Alippi, C., Webb, G

    Jin, M., Koh, H. Y., Wen, Q., Zambon, D., Alippi, C., Webb, G. I., King, I., and Pan, S. A survey on graph neural networks for time series: Forecasting, classification, imputation, and anomaly detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024 a

    2024

  8. [16]

    Y., Shi, X., Chen, P.-Y., Liang, Y., Li, Y.-F., Pan, S., et al

    Jin, M., Wang, S., Ma, L., Chu, Z., Zhang, J. Y., Shi, X., Chen, P.-Y., Liang, Y., Li, Y.-F., Pan, S., et al. Time-llm: Time series forecasting by reprogramming large language models. In International Conference on Learning Representations, 2024 b

    2024

  9. [17]

    Position: What can large language models tell us about time series analysis

    Jin, M., Zhang, Y., Chen, W., Zhang, K., Liang, Y., Yang, B., Wang, J., Pan, S., and Wen, Q. Position: What can large language models tell us about time series analysis. In FInternational Conference on Machine Learning, 2024 c

    2024

  10. [18]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Li, J., Li, D., Xiong, C., and Hoi, S. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In International conference on machine learning, pp.\ 12888--12900. PMLR, 2022

    2022

  11. [19]

    Rouge: A package for automatic evaluation of summaries

    Lin, C.-Y. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out: Proceedings of the ACL-04 workshop, volume 8, pp.\ 74--81, 2004

    2004

  12. [20]

    B., Sharma, M., Cui, J., Wen, Q., Zhang, C., et al

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

    2024

  13. [21]

    itransformer: Inverted transformers are effective for time series forecasting

    Liu, Y., Hu, T., Zhang, H., Wu, H., Wang, S., Ma, L., and Long, M. itransformer: Inverted transformers are effective for time series forecasting. In International Conference on Learning Representations, 2024 b

    2024

  14. [22]

    Autotimes: Autoregressive time series forecasters via large language models

    Liu, Y., Qin, G., Huang, X., Wang, J., and Long, M. Autotimes: Autoregressive time series forecasters via large language models. In Advances in Neural Information Processing Systems, 2024 c

    2024

  15. [23]

    Timer : Generative pre-trained transformers are large time series models

    Liu, Y., Zhang, H., Li, C., Huang, X., Wang, J., and Long, M. Timer : Generative pre-trained transformers are large time series models. In International Conference on Machine Learning, 2024 d

    2024

  16. [24]

    Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks

    Lu, J., Batra, D., Parikh, D., and Lee, S. Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks. Advances in neural information processing systems, 32, 2019

    2019

  17. [25]

    Y., Chen, B., Williamson, D

    Lu, M. Y., Chen, B., Williamson, D. F. K., Chen, R. J., Zhao, M., Chow, A. K., Ikemura, K., Kim, A., Pouli, D., Patel, A., et al. A multimodal generative ai copilot for human pathology. Nature, 634 0 (8033): 0 466--473, 2024

    2024

  18. [26]

    Nguyen, N., Sinthong, P., and Kalagnanam, J

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

    2023

  19. [27]

    Gpt-4 technical report, 2024

    OpenAI. Gpt-4 technical report, 2024. URL https://openai.com/research/gpt-4

    2024

  20. [28]

    N., Carpov, D., Chapados, N., and Bengio, Y

    Oreshkin, B. N., Carpov, D., Chapados, N., and Bengio, Y. N-beats: Neural basis expansion analysis for interpretable time series forecasting. In International Conference on Learning Representations, 2022

    2022

  21. [29]

    Bleu: a method for automatic evaluation of machine translation

    Papineni, K., Roukos, S., Ward, T., and Zhu, W.-J. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics (ACL), pp.\ 311--318. ACL, 2002

    2002

  22. [30]

    Squad: 100,000+ questions for machine comprehension of text

    Rajpurkar, P. Squad: 100,000+ questions for machine comprehension of text. arXiv preprint arXiv:1606.05250, 2016

    2016 arXiv

  23. [31]

    M., Xing, E., Yang, M.-H., and Khan, F

    Rasheed, H., Maaz, M., Shaji, S., Shaker, A., Khan, S., Cholakkal, H., Anwer, R. M., Xing, E., Yang, M.-H., and Khan, F. S. GLAMM : Pixel grounding large multimodal model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 13009--13018, 2024

    2024

  24. [32]

    J., and Manning, C

    See, A., Liu, P. J., and Manning, C. D. Get to the point: Summarization with pointer-generator networks. arXiv preprint arXiv:1704.04368, 2017

    2017 arXiv

  25. [33]

    H., Stoffer, D

    Shumway, R. H., Stoffer, D. S., Shumway, R. H., and Stoffer, D. S. Arima models. Time series analysis and its applications: with R examples, pp.\ 75--163, 2017

    2017

  26. [34]

    Generative multimodal models are in-context learners

    Sun, Q., Cui, Y., Zhang, X., Zhang, F., Yu, Q., Wang, Y., Rao, Y., Liu, J., Huang, T., and Wang, X. Generative multimodal models are in-context learners. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 14398--14409, 2024

    2024

  27. [35]

    and Bansal, M

    Tan, H. and Bansal, M. Lxmert: Learning cross-modality encoder representations from transformers. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), 2019

    2019

  28. [36]

    M., Hauth, A., Millican, K., et al

    Team, G., Anil, R., Borgeaud, S., Alayrac, J.-B., Yu, J., Soricut, R., Schalkwyk, J., Dai, A. M., Hauth, A., Millican, K., et al. Gemini: a family of highly capable multimodal models. arXiv preprint arXiv:2312.11805, 2023

    2023 arXiv

  29. [37]

    Llama: Open and efficient foundation language models

    Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozi \`e re, B., Goyal, N., Hambro, E., Azhar, F., Rodriguez, A., Joulin, A., Grave, E., and Chintala, S. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023

    2023 arXiv

  30. [38]

    Attention is all you need

    Vaswani, A. Attention is all you need. Advances in Neural Information Processing Systems, 2017

    2017

  31. [39]

    Chattime: A unified multimodal time series foundation model bridging numerical and textual data

    Wang, C., Qi, Q., Wang, J., Sun, H., Zhuang, Z., Wu, J., Zhang, L., and Liao, J. Chattime: A unified multimodal time series foundation model bridging numerical and textual data. arXiv preprint arXiv:2412.11376, 2024 a

    2024 arXiv

  32. [40]

    From news to forecast: Integrating event analysis in llm-based time series forecasting with reflection

    Wang, X., Feng, M., Qiu, J., Gu, J., and Zhao, J. From news to forecast: Integrating event analysis in llm-based time series forecasting with reflection. arXiv preprint arXiv:2409.17515, 2024 b

    2024 arXiv

  33. [41]

    Establishment of super sonic inlet flow pattern monitoring system: A workflow

    Wang, Y., Han, Z., Zhao, Y., Wu, H., Tan, H.-J., Zhang, Y., and Li, Y. Establishment of super sonic inlet flow pattern monitoring system: A workflow. Aerospace Science and Technology, 120: 0 107297, January 2022. ISSN 1270-9638

    2022

  34. [42]

    Self-supervised health representation decomposition based on contrast learning

    Wang, Y., Shen, L., Zhang, Y., Li, Y., Zhang, R., and Yang, Y. Self-supervised health representation decomposition based on contrast learning. Reliability Engineering & System Safety, 239: 0 109455, 2023

    2023

  35. [43]

    Incorporating prior knowledge into self-supervised representation learning for long PHM signal

    Wang, Y., Li, Y., Zhang, Y., Lei, J., Yu, Y., Zhang, T., Yang, Y., and Zhao, H. Incorporating prior knowledge into self-supervised representation learning for long PHM signal. Reliability Engineering & System Safety, 241: 0 109602, January 2024 c . ISSN 09518320

    2024

  36. [44]

    Leveraging large self-supervised time-series models for transferable diagnosis in cross-aircraft type bleed air system

    Wang, Y., Lei, P., Wang, X., Jiang, L., Xuan, L., Cheng, W., Zhao, H., and Li, Y. Leveraging large self-supervised time-series models for transferable diagnosis in cross-aircraft type bleed air system. Advanced Engineering Informatics, 65: 0 103275, 2025 a . ISSN 1474-0346. do...

    2025

  37. [45]

    Data-driven dynamic health index construction for diagnosis and prognosis of engine bleed air system

    Wang, Y., Zhao, H., Cheng, W., Zhang, Y., Jia, L., and Li, Y. Data-driven dynamic health index construction for diagnosis and prognosis of engine bleed air system. Aerospace Systems, 8 0 (1): 0 149--161, 2025 b

    2025

  38. [46]

    Emergent abilities of large language models

    Wei, J., Tay, Y., Bommasani, R., Raffel, C., Zoph, B., Borgeaud, S., Yogatama, D., Bosma, M., Zhou, D., Metzler, D., et al. Emergent abilities of large language models. Transactions on Machine Learning Research, 2022

    2022

  39. [47]

    Interpretable weather forecasting for worldwide stations with a unified deep model

    Wu, H., Zhou, H., Long, M., and Wang, J. Interpretable weather forecasting for worldwide stations with a unified deep model. Nature Machine Intelligence, 5 0 (6): 0 602--611, 2023

    2023

  40. [48]

    Qwen2.5 technical report

    Yang, A., Yang, B., Zhang, B., Hui, B., Zheng, B., Yu, B., Li, C., Liu, D., Huang, F., Wei, H., Lin, H., Yang, J., Tu, J., Zhang, J., Yang, J., Yang, J., Zhou, J., Lin, J., Dang, K., Lu, K., Bao, K., Yang, K., Yu, L., Li, M., Xue, M., Zhang, P., Zhu, Q., Men, R., Lin, R., Li, ...

    2024 arXiv

  41. [49]

    Learning topology-agnostic eeg representations with geometry-aware modeling

    Yi, K., Wang, Y., Ren, K., and Li, D. Learning topology-agnostic eeg representations with geometry-aware modeling. Advances in Neural Information Processing Systems, 36, 2024

    2024

  42. [50]

    Coca: Contrastive captioners are image-text foundation models

    Yu, J., Wang, Z., Vasudevan, V., Yeung, L., Seyedhosseini, M., and Wu, Y. Coca: Contrastive captioners are image-text foundation models. Transactions on Machine Learning Research, 2022. ISSN 2835-8856

    2022

  43. [51]

    TS2Vec : Towards universal representation of time series

    Yue, Z., Wang, Y., Duan, J., Yang, T., Huang, C., Tong, Y., and Xu, B. TS2Vec : Towards universal representation of time series. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pp.\ 8980--8987, 2022

    2022

  44. [52]

    and Yan, J

    Zhang, Y. and Yan, J. Crossformer : Transformer utilizing cross-dimension dependency for multivariate time series forecasting. In International Conference on Learning Representations, 2023

    2023

  45. [53]

    I., and Wang, J

    Zhang, Y., Long, M., Chen, K., Xing, L., Jin, R., Jordan, M. I., and Wang, J. Skilful nowcasting of extreme precipitation with NowcastNet . Nature, 619 0 (7970): 0 526--532, 2023

    2023

  46. [54]

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

    Zhou, H., Zhang, S., Peng, J., Zhang, S., Li, J., Xiong, H., and Zhang, W. Informer: Beyond efficient transformer for long sequence time-series forecasting. In Proceedings of the AAAI conference on artificial intelligence, volume 35, pp.\ 11106--11115, 2021

    2021

  47. [55]

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

    Zhou, T., Ma, Z., Wen, Q., Wang, X., Sun, L., and Jin, R. FEDformer : Frequency enhanced decomposed transformer for long-term series forecasting. In International Conference on Machine Learning, 2022

    2022

  48. [56]

    Y., Yuhao Cui, D

    Zhou Yu, J. Y., Yuhao Cui, D. T., and Tian, Q. Deep modular co-attention networks for visual question answering. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 6281--6290, 2019

    2019

  49. [57]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.