Pith. sign in

REVIEW 3 major objections 4 minor 60 references

Decoupling Perception from Description: Computation-Grounded Representation Alignment between Multivariate Time Series and Language

T0 review · 3 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read A 4B-parameter model called CGTime, trained to verbalize deterministically computed statistics rather than LLM-written descriptions, outperforms far larger general-purpose models at retrieving multivariate time-series facts.

desk verdict Genuinely new supervision idea, thoroughly executed, but the headline comparison is measured on benchmarks built from the same 169-metric pipeline that provides training and reward, so the multivariate advantage is not yet established. read the letter →

arxiv 2608.05238 v1 pith:ZZ3POGMX submitted 2026-08-05 cs.LG

Xinran Feng , Yi Xie , Chao Zhang , Ruikun Li , Wanyun Ling , Ziyue Li , Chenxi Liu This is my paper · ORCID
classification cs.LG
keywords computation-groundedsupervisionmultivariatetimeseriestime-serieslanguagemodelsverifiablerewardrepresentationalignmentPCAsummarizationstatisticalfactretrievalmetricQA
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

Time-series–language models are usually trained on descriptions written by LLMs that read the series, so the labels cannot be more perceptually skilled than the model being trained. This paper argues that supervision should decouple perception from description: deterministic code computes a fixed library of 169 statistics from real multivariate series, and an LLM only verbalizes those precomputed facts. The result is CGTime, a 4B-parameter model whose supervised fine-tuning and reinforcement-learning reward are both anchored to the same computed facts. On a held-out multivariate Metric-QA set, CGTime scores 0.283 versus 0.203 for GPT-5.4-nano and 0.173 for GPT-4o-mini, with the advantage surviving Holm-corrected paired significance tests. If this holds, it means multivariate statistical fact retrieval can be instilled cheaply by computation-grounded supervision rather than by scaling model size.

What carries the argument

The central mechanism is the metric-function library $\mathcal{F}=\{f_a\}$ with $A=169$ deterministic statistic functions, each computing one marginal, cross-channel, or system-level property from a masked multivariate series. To avoid the $O(K^2)$ blow-up of describing every channel pair, PCA is applied to extract $r\ll K$ principal components and the description covers each channel's correlation with retained components plus selected lead-lag relations, reducing complexity to $O(rK)$. The same computed facts are dual-use: they supervise SFT targets across a four-level curriculum and define a Gaussian-kernel verifiable reward in GRPO, whose factual component is computed by a rule-based extractor matching generated claims against $M_i$. The architecture is a frozen MOMENT encoder, a trainable channel–patch alignment module that compresses into temporal and channel-summary tokens, a scale encoder adding per-channel mean and standard-deviation tokens, and an unfrozen Qwen3-4B language model.

What would settle it

Run CGTime on an external benchmark whose questions target statistical properties outside the 169-metric inventory, or whose ground truth comes from human expert annotation on unseen domains; if its score falls to parity with GPT-4o-mini or GPT-5.4-nano, the claimed advantage is benchmark familiarity rather than decoupled perception.

Watch

Extended reading notes

Core claim

The paper's central claim is that the reliability–realism–scalability trilemma of time-series–language supervision can be resolved by splitting the two roles that LLM-written labels conflate. Perception is the part LLMs do badly, so CGTime assigns it to deterministic computation: a library of 169 metric functions computes verifiable statistics from each real series, including PCA-based summaries of cross-channel structure, lead-lag relations, and risk indicators. Description is what the language model actually learns: it is trained to verbalize those precomputed facts, with the same statistics serving as ground truth for supervised fine-tuning and as a verifiable reward during GRPO. The author's evidence is that the 4B model outperforms far larger general-purpose and specialist baselines on the multivariate metric-QA benchmark and writes captions with more accurate and more numerous numerical claims. On the paper's own terms, this establishes that computation-grounded supervision instills multivariate statistical fact retrieval into a small model better than zero-shot prompting of large LLMs.

Load-bearing premise

The load-bearing premise is that the internal Metric-QA and Captioning benchmarks, whose questions and scoring are anchored to the same computed facts used for training and reward, are a valid and unbiased measure of multivariate time-series understanding; if they over-reward task familiarity, the headline advantage over zero-shot baselines does not establish general understanding.

Editorial extensions

If this is right

  • Supervision for multivariate time-series language models can be generated automatically at scale, because deterministic code replaces human or LLM labeling of every channel pair.
  • The PCA reduction from $O(K^2)$ to $O(rK)$ description terms makes cross-channel structure describable without enumerating all pairs, supporting models on high-dimensional series.
  • Using the same computed facts for SFT targets and RL rewards gives a consistent, verifiable optimization signal, so numerical claims in generated text can be audited against the source series.
  • The model's multivariate score holds at 0.283 versus an overall 0.288, while the strongest baselines drop from their overall scores, indicating the gain concentrates where cross-channel structure matters.
  • Some statistical families remain weak: direct level and scale readout scores 0.137 on forced-answer Metric-QA, so exact raw-value recovery is the identified bottleneck.

Reading between the lines

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

  • If the benchmark anchors evaluation to the same facts used in training, the headline advantage may partly reflect task familiarity; a human-annotated, out-of-distribution statistic benchmark would be needed to confirm general understanding.
  • The decoupling principle transfers to other modalities where perception is hard but verification is cheap, such as audio, video, or sensor streams, provided a deterministic fact library exists.
  • The Direct-readout bottleneck suggests a testable extension: adding a raw-value readout path or explicit value tokens could close the gap on exact level and scale statistics without discarding the compact representation.
  • The 0.186 caption recall (versus 0.018 to 0.141 for external models) suggests that the density of verifiable claims, not just their correctness, is a trainable property that could be pushed further through reward shaping on coverage.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes CGTime, a 4B-parameter time-series–language model trained by decoupling perception from description: deterministic code computes 169 statistical properties from real multivariate series, an LLM verbalizes those precomputed facts, and the same computed facts serve as supervised-training targets, RL rewards, and evaluation references. The authors claim that CGTime outperforms much larger general-purpose and specialist models on multivariate time-series understanding, with the headline result being a Metric-QA mean fact score of 0.283 on the multivariate held-out set versus 0.203 for GPT-5.4-nano and 0.173 for GPT-4o-mini, with Holm-corrected paired significance. They also report higher caption precision/recall and competitive TSQA accuracy. The paper includes detailed appendices on data construction, prompts, scoring, significance testing, ablations, and reproducibility artifacts.

Significance. If the empirical claim held on a genuinely unseen benchmark, the decoupling principle would be a valuable contribution to time-series–language alignment: replacing LLM-perceived labels with deterministic, verifiable statistics is a sensible way to escape the self-supervision trap, and the paper's open pipeline, reproducible scoring, and ablation comparing computed versus GPT-perceived supervision (Appendix M) are concrete strengths. However, the central evidence is currently confined to benchmarks constructed from the same 169-metric pipeline and scored with the same Gaussian kernel used as the training reward, and the only external benchmark (TSQA) is itself included in training. The claimed general multivariate understanding therefore is not yet established beyond in-distribution optimization. The paper is transparent about this anchoring (Appendices G and H explicitly state that M_i grounds training and evaluation), which is commendable, but transparency does not remove the need for a valid transfer test.

major comments (3)
  1. [§3.2, App. L, Tables 17–18] The evaluation is circular with respect to the training objective. Metric-QA and Captioning are generated from the same 169-metric pipeline that produced training labels and RL rewards, and the evaluation scorer is the same Gaussian kernel used as the training reward (Eq. (2) and Eq. (4), with evaluation fixing m=0.5 while training uses split-specific multipliers). The paper's own protocol states that 'Metric-QA and Captioning use the computed statistics in M_i for supervision, RL rewards, and evaluation' (App. G.1). The only external benchmark, TSQA, is explicitly included in the training mixture during Joint-SFT and Joint-GRPO (App. D.1), and its evaluation split is disjoint from training only at the series-signature level, not at the level of question construction or metric definitions. Consequently, the headline multivariate advantage (Table 2) may reflect optimization toward the exact metric inventory, tolerances, and scoring function, rather than general multivariate time-series understanding. A true transfer test is required: for example, evaluating on a benchmark whose question set, metric definitions, and scoring function were entirely unseen during training, or training without TSQA and then evaluating on TSQA. Until such a test is provided, the central claim should be read as in-distribution benchmark performance, not general capability.
  2. [§1 vs Table 10] The statistic-family analysis weakens the interpretation of the multivariate advantage as evidence of general understanding. Table 17 shows that CGTime scores 0.137 on Direct level/scale readout questions, far below GPT-5.4-nano's 0.590, while its strong performance is concentrated in the Joint relational/system family (0.283), which coincides with the PCA, synchrony, and lead-lag statistics that are emphasized in training and rewarded during RL. Because Direct and Marginal questions appear only in the univariate split and Joint questions only in the multivariate split (Table 16), the family comparison is nested with variable count, and the Joint-family advantage is not separable from the multivariate training emphasis. This pattern is exactly what would be expected if the model had learned to reproduce the specific computed statistics it was rewarded to emit, rather than acquiring a general multivariate understanding.
  3. [§2.3, §3.1, App. E/F] There is an internal inconsistency in the exclusion of causal links. Section 1 states: 'We exclude causal links, which require assumptions beyond observational data.' Yet the metric inventory in Table 10 includes granger_xy_pvalue and granger_yx_pvalue, which are Granger-causality test p-values. If these metrics are computed, used in training, or evaluated, the stated exclusion is violated; if they are not used, the discrepancy should be clarified in the inventory and in the main text. This matters because the 'no causal assumptions' claim is part of the paper's methodological framing.
minor comments (4)
  1. [Table 3] The term 'held-out benchmark' is used for Metric-QA and Captioning, but held-out refers only to time-series signatures not seen in training; the question templates, metric definitions, extraction rules, and scoring function are the same as those used to generate training labels and RL rewards. Consider renaming this to 'series-disjoint internal benchmark' and explicitly stating the remaining alignment with the training pipeline in the main text.
  2. [App. I.4, App. H.1] The TSQA comparison with PATRA is footnoted as not being head-to-head because PATRA uses a different split and evaluation protocol, but the main text says CGTime is 'numerically close to PATRA’s strongest reported results.' This phrasing invites a direct comparison that the protocol does not support; please add an explicit caveat in the main text or move the PATRA comparison entirely to the appendix.
  3. [§2.6, App. G.1] The captioning extractor is the same rule-based extractor used to compute the RL reward, and the caption score uses the same Gaussian kernel. This means caption scores partially reflect how well a model's phrasing matches the extractor's patterns, not only the factual content. A short discussion of this extraction-conditioned nature in the main text would improve interpretability.
  4. [§2.6, App. G.1] The paper states in Appendix G.1 that 'Metric-QA and Captioning use the computed statistics in M_i for supervision, RL rewards, and evaluation.' This is transparent, but in the main text the same fact is only implicit. A prominent limitation paragraph in the main text would help readers avoid overinterpreting the internal benchmark results.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the headline comparison is an out-of-sample empirical result; the shared use of the computed metric set Mi for supervision, reward, and evaluation is a disclosed benchmark-alignment choice, not a reduction by construction.

full rationale

The paper's central claims are empirical rather than derivational, and no load-bearing step reduces to its own inputs by the paper's own equations. The Multivariate Fact Score is computed on a fixed 2,000-request held-out set whose time-series signatures are audited against training (Appendix B), and all headline comparisons are paired significance tests over those held-out items (Appendix J). The same deterministic metric library Mi is intentionally used for training supervision, RL rewards, and evaluation (§2.1, §2.6, Appendix G.1), and the same Gaussian kernel family appears in the reward (Eq. 2) and in the evaluation scorer (Eq. 4). This is benchmark alignment, not circularity: no parameter is fitted to the evaluation scores, the model must generate values for series it has not seen, and the evaluation set is not a subset of the training set. The external TSQA benchmark is also split 90/10 with an overlap audit, so its test subset is out-of-sample even though its training split is mixed into Joint-SFT and Joint-GRPO. Appendix L's statistic-family analysis is an honest empirical decomposition and even identifies a weakness in Direct level/scale readout, which counts against a story that the benchmark was engineered to flatter the model. The only notable internal inconsistency is the stated exclusion of causal links in §1 versus the presence of granger_xy_pvalue and granger_yx_pvalue in Table 10; that is a correctness or consistency flaw, not a circular argument. No self-citation chain, imported uniqueness theorem, or ansatz smuggled via citation appears. Accordingly, no specific circular step can be exhibited, and the appropriate score is 0.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claims rest on the correctness of the deterministic statistic computations, the completeness of the 169-metric inventory, and the validity of the evaluation protocol. The main free parameters are the per-metric tolerances and kernel bandwidths that define accuracy, plus thresholds for PCA, captions, and anomalies. No new entities are postulated.

free parameters (5)
  • Per-metric tolerance table c (and integer tau) = Varied by metric: 0.04-0.15 for [0,1] metrics, 0.08-0.10 for correlations, 0.15/0.20 relative for unbounded magnitudes
    These tolerances define what counts as a correct numerical answer; they are chosen by the authors and directly shape scores and RL rewards.
  • Gaussian bandwidth multiplier m = 0.5 at evaluation; 0.5/0.5/1.0 by split during Metric-QA RL training
    Controls the sharpness of the accuracy kernel; evaluation uses the same kernel as the training reward.
  • PCA retention threshold tau = Not stated precisely; subject to a 20% dimensionality floor and clipping to [1,10]
    Determines the number of retained principal components and all PCA-based statistics.
  • Caption recall threshold h and cap D_max = h thresholds kernel scores; D_max = 7
    Set the caption reward and evaluation recall denominator.
  • Mahalanobis outlier threshold = Not stated
    Defines anomaly ratios and outlier counts used across risk metrics.
assumptions (5)
  • domain assumption The deterministic code computes the 169 statistics correctly from real series
    The method and all internal ground truths depend on these calculations being error-free; no independent audit of the statistic implementations is provided.
  • ad hoc to paper The 169-metric inventory captures the meaningful statistical properties of multivariate time series
    The benchmark is defined by this inventory; coverage is assumed rather than derived from any external standard.
  • domain assumption Frozen MOMENT embeddings preserve enough information to recover the 169 statistics
    CGTime must read out exact values from compressed patch embeddings; Appendix L shows direct level/scale readout is the model's weakest family, so this assumption holds only partially.
  • domain assumption LLM verbalization of precomputed facts yields faithful natural-language captions
    The pipeline assumes the language model polishes wording without distorting numerical content; no check for distortion is reported.
  • domain assumption The rule-based extractor identifies numerical claims fairly across models
    Caption precision and recall are conditional on this extractor, which may favor models trained with the same phrasing conventions as the training captions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Decoupling Perception from Description: Computation-Grounded Representation Alignment between Multivariate Time Series and Language." pith.science (2026). https://pith.science/paper/ZZ3POGMX

@misc{pith2026260805238,
  author       = {Pith},
  title        = {Pith review of: Decoupling Perception from Description: Computation-Grounded Representation Alignment between Multivariate Time Series and Language},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZZ3POGMX}},
  note         = {Machine review of arXiv:2608.05238}
}
read the original abstract

Training multimodal models to align time series with language runs into a self-supervision trap. The usual recipe asks an LLM to read a series and write a description, so label quality is capped by the perceptual skill the model is supposed to learn. The data can never teach more than the labeler already knows. A second gap makes this worse: most datasets use a single variable, but the patterns that matter (cross-channel correlation, lead-lag structure, co-occurring anomalies) appear only with several variables, right where the labeling LLM's limits are most exposed. These two problems create a trilemma: existing methods are reliable, realistic, or scalable, but none achieves all three. We resolve this by decoupling perception from description. Deterministic code computes a set of statistics from real, open-source multivariate series; the LLM verbalizes those precomputed facts. Perception, which LLMs do poorly, is handled by computation, while the LLM handles expression. This produces CGTime, our 4B-parameter computation-grounded time-series-language model. CGTime outperforms far larger general-purpose models on multivariate understanding tasks: it attains the best multivariate fact score on our held-out benchmark (0.283 vs. 0.173 for GPT-4o-mini and 0.203 for GPT-5.4-nano), a gap that survives Holm-corrected paired significance tests against every baseline. It also states verifiable numerical facts in generated captions more accurately and covers a broader range of statistical properties.

Figures

Figures reproduced from arXiv: 2608.05238 by the authors.

Figure 1
Figure 1. Overview of statistics-grounded supervision. Pipelines based on LLM-generated labels ask the same model to interpret [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Computation-grounded time-series data generation. Deterministic code extracts time-domain statistics and a PCA [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. CGTime architecture and training. A frozen MOMENT encoder maps each channel to patch embeddings. The two [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Contrasting diagnoses of the same five-channel se [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Metric-QA performance profiles on the univariate and multivariate evaluation splits. Panel (a) reports per-split Gaussian [PITH_FULL_IMAGE:figures/full_fig_p026_5.png]
Figure 6
Figure 6. Figure 6: Caption factuality and support under the fixed rule-based extraction method. The area of gray and colored model [PITH_FULL_IMAGE:figures/full_fig_p028_6.png]
Figure 7
Figure 7. Figure 7: Caption Case 1: a stable low-dimensional system is contrasted with a diagnosis of structural-change risk. [PITH_FULL_IMAGE:figures/full_fig_p032_7.png]
Figure 8
Figure 8. Figure 8: Caption Case 2: synchronized fluctuations are contrasted with a diagnosis of structural fragility. [PITH_FULL_IMAGE:figures/full_fig_p035_8.png]
Figure 9
Figure 9. Figure 9: Caption Case 3: persistent common-factor structure is contrasted with inferred state transitions. [PITH_FULL_IMAGE:figures/full_fig_p038_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 34 canonical work pages

  1. [1]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  2. [2]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  3. [3]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  4. [4]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  5. [5]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    doi:10.1016/s0020-7373(84)80003-6 1984

  6. [6]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  7. [7]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  8. [8]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

Show all 60 references
  1. [9]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  2. [10]

    2023 , eprint=

    Attention Is All You Need , author=. 2023 , eprint=

    2023

  3. [11]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

  4. [12]

    Xie, Zhe and Li, Zeyan and He, Xiao and Xu, Longlong and Wen, Xidao and Zhang, Tieying and Chen, Jianjun and Shi, Rui and Pei, Dan , journal =

  5. [13]

    Kong, Yaxuan and Yang, Yiyuan and Hwang, Yoontae and Du, Wenjie and Zohren, Stefan and Wang, Zhangyang and Jin, Ming and Wen, Qingsong , booktitle =

  6. [14]

    Wang, Yilin and Lei, Peixuan and Song, Jie and Hao, Yuzhe and Chen, Tao and Zhang, Yuxuan and Jia, Lei and Li, Yuanxiang and Wei, Zhongyu , booktitle =

  7. [15]

    Guan, Tong and Meng, Zijie and Li, Dianqi and Wang, Shiyu and Yang, Chao-Han Huck and Wen, Qingsong and Liu, Zuozhu and Siniscalchi, Sabato Marco and Jin, Ming and Pan, Shirui , journal =

  8. [16]

    and Yang, Bin , journal =

    Lu, Junkai and Chen, Peng and Wu, Xingjian and Shu, Yang and Guo, Chenjuan and Jensen, Christian S. and Yang, Bin , journal =

  9. [17]

    Findings of the Association for Computational Linguistics: EMNLP 2024 , pages=

    Language models still struggle to zero-shot reason about time series , author=. Findings of the Association for Computational Linguistics: EMNLP 2024 , pages=

    2024

  10. [18]

    Computer Graphics Forum , volume =

    Bertin Was Right: An Empirical Evaluation of Indexing to Compare Multivariate Time-Series Data Using Line Plots , author =. Computer Graphics Forum , volume =. 2011 , doi =

    2011

  11. [19]

    Evaluating

    Arai, Mizuki and Ishigaki, Tatsuya and Kawarada, Masayuki and Miyao, Yusuke and Takamura, Hiroya and Kobayashi, Ichiro , booktitle =. Evaluating. 2025 , url =

    2025

  12. [20]

    Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan , booktitle =

  13. [21]

    Advances in Neural Information Processing Systems (NeurIPS) , year =

    Toolformer: Language Models Can Teach Themselves to Use Tools , author =. Advances in Neural Information Processing Systems (NeurIPS) , year =

  14. [22]

    Ding, Chris and He, Xiaofeng , booktitle =

  15. [23]

    Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence (IJCAI) , year =

    Large Language Models for Time Series: A Survey , author =. Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence (IJCAI) , year =

  16. [24]

    and Shi, Xiaoming and Chen, Pin-Yu and Liang, Yuxuan and Li, Yuan-Fang and Pan, Shirui and Wen, Qingsong , booktitle =

    Jin, Ming and Wang, Shiyu and Ma, Lintao and Chu, Zhixuan and Zhang, James Y. and Shi, Xiaoming and Chen, Pin-Yu and Liang, Yuxuan and Li, Yuan-Fang and Pan, Shirui and Wen, Qingsong , booktitle =

  17. [25]

    arXiv preprint arXiv:2403.07815 , year =

    Chronos: Learning the Language of Time Series , author =. arXiv preprint arXiv:2403.07815 , year =

    arXiv

  18. [26]

    International Conference on Machine Learning (ICML) , year =

    Unified Training of Universal Time Series Forecasting Transformers , author =. International Conference on Machine Learning (ICML) , year =

  19. [27]

    International Conference on Machine Learning (ICML) , year =

    A Decoder-Only Foundation Model for Time-Series Forecasting , author =. International Conference on Machine Learning (ICML) , year =

  20. [28]

    Goswami, Mononito and Szafer, Konrad and Choudhry, Arjun and Cai, Yifu and Li, Shuo and Dubrawski, Artur , booktitle =

  21. [29]

    arXiv preprint arXiv:2501.12948 , year =

    arXiv

  22. [30]

    Shao, Zhihong and Wang, Peiyi and Zhu, Qihao and Xu, Runxin and Song, Junxiao and Bi, Xiao and Zhang, Haowei and Zhang, Mingchuan and Li, Y. K. and Wu, Y. and Guo, Daya , journal =

  23. [31]

    Principal Component Analysis , author =

  24. [32]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume =

    Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting , author =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =. 2021 , doi =

    2021

  25. [33]

    Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks , volume =

    Monash Time Series Forecasting Archive , author =. Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks , volume =

  26. [34]

    International conference on learning representations , volume=

    Time-moe: Billion-scale time series foundation models with mixture of experts , author=. International conference on learning representations , volume=

  27. [35]

    International Journal of Forecasting , volume=

    The M4 Competition: 100,000 time series and 61 forecasting methods , author=. International Journal of Forecasting , volume=. 2020 , publisher=

    2020

  28. [36]

    2015 , howpublished =

    Trindade, Artur , title =. 2015 , howpublished =

    2015

  29. [37]

    Knowledge-Based Systems , volume=

    MvTS-library: An open library for deep multivariate time series forecasting , author=. Knowledge-Based Systems , volume=. 2024 , publisher=

    2024

  30. [38]

    Hersbach, Hans and Bell, Bill and Berrisford, Paul and Hirahara, Shoji and Hor. The. Quarterly Journal of the Royal Meteorological Society , volume =. 2020 , doi =

    2020

  31. [39]

    and Ng, Serena , journal =

    McCracken, Michael W. and Ng, Serena , journal =. 2016 , doi =

    2016

  32. [40]

    Sensors and Actuators B: Chemical , volume =

    On Field Calibration of an Electronic Nose for Benzene Estimation in an Urban Pollution Monitoring Scenario , author =. Sensors and Actuators B: Chemical , volume =. 2008 , doi =

    2008

  33. [41]

    2008 , howpublished =

    Vito, Saverio , title =. 2008 , howpublished =

    2008

  34. [42]

    Energy and Buildings , volume =

    Data Driven Prediction Models of Energy Use of Appliances in a Low-Energy House , author =. Energy and Buildings , volume =. 2017 , doi =

    2017

  35. [43]

    2017 , howpublished =

    Candanedo, Luis , title =. 2017 , howpublished =

    2017

  36. [44]

    2018 6th International Renewable and Sustainable Energy Conference (IRSEC) , year =

    Comparison of Machine Learning Algorithms for the Power Consumption Prediction: Case Study of Tetouan City , author =. 2018 6th International Renewable and Sustainable Energy Conference (IRSEC) , year =

    2018

  37. [45]

    2018 , howpublished =

    Salam, Abdulwahed and El Hibaoui, Abdelaaziz , title =. 2018 , howpublished =

    2018

  38. [46]

    2019 , howpublished =

    Hogue, John , title =. 2019 , howpublished =

    2019

  39. [47]

    2025 , note =

    Wang, Chengsen and Qi, Qi and Wang, Jingyu and Sun, Haifeng and Zhuang, Zirui and Wu, Jinming and Zhang, Lei and Liao, Jianxin , booktitle =. 2025 , note =

    2025

  40. [48]

    arXiv preprint arXiv:2501.01832 , year =

    Time Series Language Model for Descriptive Caption Generation , author =. arXiv preprint arXiv:2501.01832 , year =

    arXiv

  41. [49]

    Bayan and Nguyen, Nam and Hartvigsen, Tom , journal =

    Sen, Medhasweta and Gottesman, Zachary and Qiu, Jiaxing and Bruss, C. Bayan and Nguyen, Nam and Hartvigsen, Tom , journal =

  42. [50]

    2024 , note =

    Cai, Yifu and Choudhry, Arjun and Goswami, Mononito and Dubrawski, Artur , journal =. 2024 , note =

    2024

  43. [51]

    2026 , note =

    Yu, Fangxu and Guo, Xingang and Yuan, Lingzhi and Kang, Haoqiang and Zhao, Hongyu and Qin, Lianhui and Huang, Furong and Hu, Bin and Zhou, Tianyi , booktitle =. 2026 , note =

    2026

  44. [52]

    arXiv preprint arXiv:2110.14168 , year =

    Training Verifiers to Solve Math Word Problems , author =. arXiv preprint arXiv:2110.14168 , year =

    arXiv

  45. [53]

    arXiv preprint arXiv:2305.20050 , year =

    Let's Verify Step by Step , author =. arXiv preprint arXiv:2305.20050 , year =

    arXiv

  46. [54]

    arXiv preprint arXiv:2211.14275 , year =

    Solving Math Word Problems with Process- and Outcome-Based Feedback , author =. arXiv preprint arXiv:2211.14275 , year =

    arXiv

  47. [55]

    , journal =

    Zelikman, Eric and Wu, Yuhuai and Mu, Jesse and Goodman, Noah D. , journal =

  48. [56]

    arXiv preprint arXiv:2509.08827 , year =

    A Survey of Reinforcement Learning for Large Reasoning Models , author =. arXiv preprint arXiv:2509.08827 , year =

    arXiv

  49. [57]

    2508.10925 , archivePrefix =

    gpt-oss-120b & gpt-oss-20b Model Card , year =. 2508.10925 , archivePrefix =

    arXiv

  50. [58]

    2410.21276 , archivePrefix =

    GPT-4o System Card , year =. 2410.21276 , archivePrefix =

    arXiv

  51. [59]

    arXiv preprint arXiv:2505.09388 , year=

    Qwen3 technical report , author=. arXiv preprint arXiv:2505.09388 , year=

    arXiv

  52. [60]

    5 Technical Report , author=

    Qwen2. 5 Technical Report , author=. arXiv preprint arXiv:2412.15115 , year=

    arXiv

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.