TimeDP: Learning to Generate Multi-Domain Time Series with Domain Prompts

Chang Xu; Jiang Bian; Wu-Jun Li; Yueying Wu; Yu-Hao Huang

REVIEW 4 major objections 6 minor 1 cited by

TimeDP: Learning to Generate Multi-Domain Time Series with Domain Prompts

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A single diffusion model, conditioned on few-shot 'domain prompts' built from a frozen random dictionary of time-series prototypes, generates high-quality data for trained domains and adapts to unseen domains without fine-tuning.

desk verdict A useful multi-domain time series diffusion method with credible in-domain results, but the unseen-domain generalization claim runs ahead of the evidence. read the letter →

arxiv 2501.05403 v1 pith:IJZWOMP2 submitted 2025-01-09 cs.LG cs.AI

Yu-Hao Huang , Chang Xu , Yueying Wu , Wu-Jun Li , Jiang Bian This is my paper · ORCID

classification cs.LGcs.AI

keywords multi-domaintimeseriesgenerationdiffusionmodelsdomainpromptsprototypeassignmentfew-shotunseenprototypesconditional

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 proposes TimeDP, a single diffusion model that generates time series for many domains at once without domain labels or text descriptions. Instead of conditioning on a class or a natural-language prompt, TimeDP learns to construct a 'domain prompt' from a few example sequences of the target domain: a small network assigns weights to a fixed set of 16 random orthogonal prototype vectors, and those weights steer the denoising process through cross-attention. The paper claims this suffices for best-in-class in-domain generation on 12 real-world datasets and, more strikingly, for few-shot generation on domains never seen in training (Stock and Web), outperforming both unconditional baselines and baselines fine-tuned on the same few shots. If true, this offers a label-free and text-free route to multi-domain and zero-shot time-series synthesis, which matters for data augmentation, privacy-preserving synthetic data, and financial or medical scenario simulation.

What carries the argument

The load-bearing objects are the time-series semantic prototypes and the prototype assignment module (PAM). Prototypes are $N_p=16$ vectors $P \in \mathbb{R}^{N_p \times d}$, initialized as random orthogonal vectors and frozen; each vector is meant to encode an elementary time-series feature such as trend or seasonality, acting like a 'word' in a shared dictionary. PAM is a small network $\varphi$ that maps an input sequence $x$ to a weight vector $m = \varphi(x)$, with negative entries set to zero for sparsity. The weights $m$ are added to the scaled dot-product attention scores in each cross-attention layer of the denoising U-Net, so the predicted noise $\hat{\epsilon} = \epsilon_{\theta,P}(x_n, n, m)$ is conditioned on the assigned prototypes. At sampling time, the same PAM is applied to $K$ few-shot samples of the target domain to form the domain prompt $m_{D_i}=\{m^i_1,\dots,m^i_K\}$, and the model generates by repeated denoising conditioned on these prompts.

What would settle it

Hold out a domain whose time series contain frequency components absent from the twelve training datasets—for example, a synthetic chirp with a time-varying frequency, or a series of aperiodic spikes. Give TimeDP 10 or 100 samples from that domain and compute the MMD and K-L divergence between its generated samples and the true test set, alongside the same metrics for the model's own unconditional output. If the prompted generation is not closer to the target distribution than the unconditional output, or if increasing the sample count from 3 to 100 does not improve fidelity, the claim that frozen random prototypes can express unseen domains is falsified.

Watch

Extended reading notes

Core claim

TimeDP's central claim is that a small set of frozen random orthogonal vectors can act as a universal time-series 'dictionary,' and that a learned prototype assignment module can express any domain—trained or unseen—as a weighted combination of these vectors. These weights, called domain prompts, are fed into a U-Net diffusion model by biasing the cross-attention logits between the noisy latent and the prototype keys and values. During sampling, K few-shot samples from the target domain are mapped through the assignment network to obtain a prompt population, and the diffusion model generates new sequences conditioned on it. On 12 datasets across energy, transport, nature, and economic domains, TimeDP reports the best Maximum Mean Discrepancy and K-L divergence on most datasets, beating a class-conditioned vector-quantized baseline without using labels; on the unseen Stock and Web datasets it reports the best MMD, K-L, and marginal-distribution scores with 3, 10, or 100 prompt samples, improving as the few-shot count grows, while fine-tuned baselines do not consistently improve.

Load-bearing premise

The method assumes that 16 fixed, randomly chosen and frozen vectors can represent the temporal structure of any domain, including domains never seen in training, and that the assignment network trained only on source domains will map a few samples from any new domain to a meaningful prompt; if a new domain's patterns cannot be expressed by those vectors, the prompt carries no useful information and generation falls back to unconditional output.

Editorial extensions

If this is right

One trained TimeDP model can replace per-dataset generators: the same weights serve every domain, and the only per-domain input is a few dozen sample sequences.
Unseen-domain generation needs no fine-tuning: a handful of samples defines the prompt, and quality improves as the number of samples grows from 3 to 100.
Domain labels and natural-language descriptions become unnecessary for conditional time-series generation, removing a bottleneck for new or evolving domains whose patterns are hard to describe.
The prototype dictionary provides an interpretable control axis: prompting with one-hot weights yields distinct seasonal or trend patterns, so a user can steer generation in a human-readable way.
Because the prompt is a distribution over a shared basis, domains are explicitly related rather than treated as independent, which the paper argues helps the model capture inter-domain similarity.

Reading between the lines

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

Not explored in the paper: learning the prototype vectors during training instead of freezing them; if random orthogonal vectors already work, a learned, updated basis might cover a wider range of temporal patterns and improve unseen-domain transfer further.
The prompting mechanism is modality-agnostic in principle: any signal that can be decomposed into a small set of elementary patterns (images, audio, spatiotemporal data) could reuse the same few-shot prompt-extraction idea.
The ablation shows that removing PAM or the prompt barely changes K-L divergence but sharply worsens MMD, suggesting the prompt mainly preserves sequential dependence rather than marginals; a direct check of autocorrelation or spectral fidelity of generated sequences would test this interpretation.
The paper's unseen-domain experiments use only two target datasets; a stronger probe would be to hold out entire categories (for instance, all economic domains) and test whether prompts still separate them, which would clarify whether the frozen basis is truly universal or merely sufficient for these data.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

TimeDP is a solid, useful method paper. The new thing is real: conditioning a diffusion U-Net on prototype-assignment weights extracted from a few real samples, with no domain labels and no text. That combination — frozen random prototype dictionary, learned assignment network, cross-attention conditioning — is not something I've seen in the cited prior work. The empirical work is also more thorough than the average generation paper: 12 datasets, four sequence lengths, 5 seeds, several baselines, and ablations. On the in-domain benchmarks, TimeDP beats the class-conditional TimeVQVAE-C on most datasets, and the ablation study shows both the PAM and the prompt conditioning matter. The code is available. That part holds up.

The soft spots are mostly about framing and one missing definition. The "unseen domain" claim is broader than the evidence. Only two held-out datasets are tested, Stock and Web, and both are plausibly close to Exchange and NN5 in the training set. The paper asserts that the frozen random prototypes can represent any new domain and that the source-trained assignment network will map few-shot samples to useful prompts, but there is no mechanism guaranteeing that. If a genuinely novel domain produced assignments outside the training manifold, generation would degrade toward unconditional output. A claim like "strong unseen-domain capability on two held-out datasets" would be accurate; "for any unseen dataset" is not supported.

Second, Algorithm 1 has an undefined term: "Randomly set P as unconditional identifier p_u." I could guess it means randomly dropping the condition, but the exact training objective is not fully specified in the text. That is a reproducibility gap.

Third, the paper repeatedly calls the prototypes "learned" and refers to "time series semantic prototypes," but the prototype vectors are randomly initialized and frozen. What is actually learned is the assignment network and the attention projections. The visualization in Figure 2 is nice and shows those projections can extract meaningful structure, but the paper should be precise about what is learned versus fixed. This is a wording problem, not a fatal flaw.

Minor: no commit hash or environment specification in the code link, so independent verification takes some effort.

Who is this for? People working on time series generation, data augmentation, or privacy-preserving release of time series will get value from it. It is a practical method paper, not a field-rearranging one. It deserves a serious referee. I would send it to peer review with a request for revision: define p_u, soften the unseen-domain claim to match the evidence, and fix the "learned prototypes" language.

Referee Report

4 major / 6 minor

Summary. TimeDP is a multi-domain time series diffusion model that conditions generation on a "domain prompt" derived from a set of randomly initialized and then frozen prototype vectors. The Prototype Assignment Module (PAM) maps each input sequence to a sparse weight vector over the prototypes; the weights are added to the cross-attention logits of a U-Net denoiser. The model is trained jointly on 12 datasets with balanced sampling; at inference, prompts are extracted from a few-shot sample of the target domain, including domains unseen in training. The paper reports in-domain generation results on 12 datasets using MMD, K-L, and MDD metrics, plus few-shot unseen-domain results on Stock and Web, and ablations on prototype count and component removal.

Significance. TimeDP addresses a real gap: multi-domain time series generation without explicit class labels or text descriptions. The core idea of using a small set of basis vectors and sparse assignments as a conditioning interface is simple and potentially useful. The in-domain benchmark is reasonably strong: the proposed model achieves the best MMD and K-L scores on most of the 12 datasets against established baselines, and the ablation shows that the conditioning mechanism gives a large improvement in MMD over the same model without prompts. The release of code and the use of five seeds are commendable. However, the paper's broader claims—that the prototypes "learn" time-series semantics and that the method generalizes to "any unseen dataset"—are not supported as stated, because the prototype vectors are frozen and the unseen-domain evidence is limited to two near-source datasets.

major comments (4)

[Semantic Prototype Module; Abstract; Conclusion]
[Algorithm 1 (Training algorithm), line 5]
[Unseen Domain Generation; Table 2]
[Table 3 (Ablation Study)]

minor comments (6)

[Algorithm 2 (Sampling with domain prompts)]
[Equation (6)]
[Contributions and Related Work]
[Experiment Settings]
[Table 2 and Appendix Figures 6/7]
[Abstract and Background]

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: few-shot domain prompts condition generation without being defined as the predicted outputs, and unseen-domain results are evaluated on non-overlapping test samples; the main weaknesses are an unsupported generalization assumption and an undefined unconditional identifier, not circularity.

full rationale

The derivation chain is self-contained. TimeDP trains a conditional DDPM with condition m = phi(x0) (Eq. 6, Eq. 10), so the training condition is computed from the target sample x0; this is self-conditioning, but it is not circular inference because at sampling the condition is extracted from K few-shot samples of the target domain (Algorithm 2 and the 'Generation with Domain Prompt' section), and the evaluation compares generated samples with a held-out test set that 'does not overlap with these samples' (Unseen Domain Generation experiments). No quantity that the paper calls a prediction is defined as a fitted parameter: the prototype matrix P is initialized with random orthogonal vectors and frozen, and the assignment network phi is trained on source domains but applied to few-shot inputs that are not the generated outputs. The 'unseen domain' capability is asserted from the premise that prototypes' 'representation ability is not restricted to the domains in training sets,' but this is an unvalidated generalization assumption supported only by Stock and Web experiments; under-support is a correctness risk, not circularity. The self-citations (Huang et al. 2024; Fan et al. 2024; Yang et al. 2024) are used for application motivation and background, not to justify the central reduction, so they are not load-bearing. One reproducibility gap, not a circular step, is Algorithm 1's 'Randomly set P as unconditional identifier pu' where pu is never defined. Overall the claimed generation quality is an empirical result against external baselines and held-out real data, not a restatement of the model's inputs.

Assumptions & free parameters 4 free parameters · 6 assumptions · 3 invented entities

The central claim rests on a set of modeling choices: the diffusion backbone is standard, but the semantic basis is a set of hand-initialized frozen random vectors, the assignment network is learned only on source domains, and the few-shot prompt construction assumes transfer. No new physical or mathematical entities are introduced; the invented terms are internal model components with no independent evidence.

free parameters (4)

Number of prototypes Np = 16
Hand-chosen; ablation tests 4, 8, 16, 32, 64 and reports that performance is stable for larger counts, but the main claim uses 16.
Few-shot prompt count K = 3, 10, 100
Chosen for unseen-domain experiments; performance improves with K, which the paper reports.
Unconditional identifier injection probability = not specified
Algorithm 1 says 'Randomly set P as unconditional identifier pu' without specifying the probability or the form of pu, so this hyperparameter is unstated.
Negative-weight masking rule = -inf in attention
Eq. (6) sets negative assignment weights to negative infinity, a hand-chosen sparsity mechanism.

assumptions (6)

standard math DDPM forward noising and reverse denoising with epsilon-parameterization
Background from Ho et al. 2020, used as the generation backbone.
domain assumption Channel-independent univariate formulation
Adopted from Nie et al. 2023 and Woo et al. 2024 to handle heterogeneous multivariate datasets by slicing into univariate sequences.
ad hoc to paper A fixed set of 16 random orthogonal prototype vectors can serve as a time series basis across all domains
Prototypes are randomly initialized and frozen; the paper assumes they represent elementary features such as trend and seasonality without learning them.
domain assumption Each time series sample corresponds to a distinct weighted allocation over prototypes, and a few samples characterize the domain
Used in the Prototype Assignment Module and in few-shot domain prompt construction; no theoretical justification beyond analogy to basis decomposition.
domain assumption The assignment network phi trained on source domains transfers to unseen domains
Unseen-domain generation maps few-shot samples from a new domain through phi, assuming the learned assignment distribution generalizes beyond training domains.
domain assumption Mixing multiple datasets during training improves per-domain generation
Core premise of domain-unified training; the paper claims it works because domains share prototype structure.

invented entities (3)

Time series semantic prototypes P
purpose: Randomly initialized, frozen vectors that act as a shared dictionary of elementary time series features for cross-domain conditioning
Internal model component; the paper's one-hot visualization is generated by the model itself and does not provide independent evidence.
Domain prompts m_Di
purpose: Set of prototype assignment vectors extracted from a few real samples to condition generation for a target domain
Internal representation; no external benchmark links a domain prompt to an independently measurable quantity.
Unconditional identifier pu
purpose: Intended to allow unconditional generation by replacing the prototype set during training; never defined in the text
Mentioned only in Algorithm 1 line 5, with no specification of its form, learning rule, or usage during sampling.

how reviews work

0 comments

Cite this review

Pith. "Pith review of TimeDP: Learning to Generate Multi-Domain Time Series with Domain Prompts." pith.science (2026). https://pith.science/paper/IJZWOMP2

@misc{pith2026250105403,
  author       = {Pith},
  title        = {Pith review of: TimeDP: Learning to Generate Multi-Domain Time Series with Domain Prompts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IJZWOMP2}},
  note         = {Machine review of arXiv:2501.05403}
}

read the original abstract

Time series generation models are crucial for applications like data augmentation and privacy preservation. Most existing time series generation models are typically designed to generate data from one specified domain. While leveraging data from other domain for better generalization is proved to work in other application areas, this approach remains challenging for time series modeling due to the large divergence in patterns among different real world time series categories. In this paper, we propose a multi-domain time series diffusion model with domain prompts, named TimeDP. In TimeDP, we utilize a time series semantic prototype module which defines time series prototypes to represent time series basis, each prototype vector serving as "word" representing some elementary time series feature. A prototype assignment module is applied to extract the extract domain specific prototype weights, for learning domain prompts as generation condition. During sampling, we extract "domain prompt" with few-shot samples from the target domain and use the domain prompts as condition to generate time series samples. Experiments demonstrate that our method outperforms baselines to provide the state-of-the-art in-domain generation quality and strong unseen domain generation capability.

Figures

Figures reproduced from arXiv: 2501.05403 by the authors.

**Figure 1.** Overview of TimeDP model. joint distribution over time steps p(x1, x2, ..., xT ) of each dataset with a separate model parameterized by θi , namely pθi (x1, x2, ..., xT ) for all x in DT i . In this paper, we explore a domain-unified setting where the mixture of M domain datasets with sequence length T is denoted by DT = SM i=1 Di and we aim to build one model for the mixed dataset parameterized by θ, namely pθ(x1, … view at source ↗

**Figure 2.** Semantic visualization of time series prototypes. [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗

**Figure 3.** Heatmap of Domain Prompts. 40 30 20 10 0 10 20 30 40 30 20 10 0 10 20 30 Electricity Solar Wind Traffic Taxi Pedestrian Air Quality Temperature Rain NN5 Fred-MD Exchange [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: T-SNE visualization of domain prompts. Domain prompt generated for each dataset are marked with the same color. [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]

**Figure 5.** Figure 5: T-SNE visualization of domain prompts. Datasets in the same domain are marked with the same color. [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]

**Figure 6.** Figure 6: 10-shot generation results of Web dataset, compared with real Web dataset samples. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]

**Figure 7.** Figure 7: 10-shot generation results of Stock dataset, compared with real Stock dataset samples. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

TarDiff: Target-Oriented Diffusion Guidance for Synthetic Electronic Health Record Time Series Generation
cs.LG 2025-04 conditional novelty 6.0 of 10

TarDiff guides diffusion-based synthetic EHR generation with a gradient-alignment signal computed from a guidance set, reporting improved downstream mortality and ICU-stay classification versus prior generative models.

Reference graph

Works this paper leans on

28 extracted references · 21 canonical work pages · cited by 1 Pith paper

[1]

, " * write output.state after.block = add.period write newline

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...
[2]

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 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...
[3]

Coletta, A.; Gopalakrishnan, S.; Borrajo, D.; and Vyetrenko, S. 2023. On the Constrained Time-Series Generation Problem. In Advances in Neural Information Processing Systems, 61048--61059

work page 2023
[4]

Coletta, A.; Prata, M.; Conti, M.; Mercanti, E.; Bartolini, N.; Moulin, A.; Vyetrenko, S.; and Balch, T. 2021. Towards realistic market simulations: a generative adversarial networks approach. In Proceedings of the ACM International Conference on AI in Finance, 1--9

work page 2021
[5]

Das, A.; Kong, W.; Sen, R.; and Zhou, Y. 2024. A decoder-only foundation model for time-series forecasting. In Proceedings of the International Conference on Machine Learning, 10148--10167

work page 2024
[6]

Desai, A.; Freeman, C.; Wang, Z.; and Beaver, I. 2021. TimeVAE: A Variational Auto-Encoder for Multivariate Time Series Generation. arXiv preprint, arXiv:2111.08095

arXiv 2021
[7]

Fan, X.; Wu, Y.; Xu, C.; Huang, Y.; Liu, W.; and Bian, J. 2024. MG-TSD: Multi-Granularity Time Series Diffusion Models with Guided Learning Process. In International Conference on Learning Representations

work page 2024
[8]

Gao, S.; Koker, T.; Queen, O.; Hartvigsen, T.; Tsiligkaridis, T.; and Zitnik, M. 2024. UniTS: Building a Unified Time Series Model. arXiv preprint, arXiv:2403.00131

arXiv 2024

Show all 28 references

[9]

Ho, J.; Jain, A.; and Abbeel, P. 2020. Denoising Diffusion Probabilistic Models. In Advances in Neural Information Processing Systems, 6840--6851

2020
[10]

Ho, J.; and Salimans, T. 2021. Classifier-Free Diffusion Guidance. In NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications

2021
[11]

Huang, Y.-H.; Xu, C.; Liu, Y.; Liu, W.; Li, W.-J.; and Bian, J. 2024. Controllable Financial Market Generation with Diffusion Guided Meta Agent. arXiv preprint, arXiv:2408.12991

2024
[12]

Jeon, J.; KIM, J.; Song, H.; Cho, S.; and Park, N. 2022. Time-series Generative Adversarial Networks. In Advances in Neural Information Processing Systems, 36999--37010

2022
[13]

Jiang, Y.; Pan, Z.; Zhang, X.; Garg, S.; Schneider, A.; Nevmyvaka, Y.; and Song, D. 2024. Empowering Time Series Analysis with Large Language Models: A Survey. In Proceedings of the International Joint Conference on Artificial Intelligence, 8095--8103

2024
[14]

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. 2024. Time-LLM: Time Series Forecasting by Reprogramming Large Language Models. In International Conference on Learning Representations

2024
[15]

F.; Bohlke-Schneider, M.; Zschiegner, J.; Wang, H.; and Wang, Y

Kollovieh, M.; Ansari, A. F.; Bohlke-Schneider, M.; Zschiegner, J.; Wang, H.; and Wang, Y. 2023 a . Predict, refine, synthesize: self-guiding diffusion models for probabilistic time series forecasting. In Advances in Neural Information Processing Systems, 28341--28364

2023
[16]

F.; Bohlke-Schneider, M.; Zschiegner, J.; Wang, H.; and Wang, Y

Kollovieh, M.; Ansari, A. F.; Bohlke-Schneider, M.; Zschiegner, J.; Wang, H.; and Wang, Y. B. 2023 b . Predict, Refine, Synthesize: Self-Guiding Diffusion Models for Probabilistic Time Series Forecasting. In Advances in Neural Information Processing Systems, 28341--28364

2023
[17]

S.; and Kersting, K

Kraus, M.; Divo, F.; Steinmann, D.; Dhami, D. S.; and Kersting, K. 2024. United We Pretrain, Divided We Fail! Representation Learning for Time Series by Pretraining on 75 Datasets at Once. arXiv preprint, arXiv:2402.15404

2024 arXiv
[18]

Lee, D.; Malacarne, S.; and Aune, E. 2023. Vector Quantized Time Series Generation with a Bidirectional Prior Model. In Proceedings of the International Conference on Artificial Intelligence and Statistics, 7665--7693

2023
[19]

Li, H.; Yu, S.; and Pr \' ncipe, J. C. 2023. Causal Recurrent Variational Autoencoder for Medical Time Series Generation. In Williams, B.; Chen, Y.; and Neville, J., eds., Proceedings of the AAAI Conference on Artificial Intelligence , 8562--8570

2023
[20]

Ni, Z.; Yu, H.; Liu, S.; Li, J.; and Lin, W. 2023. Basisformer: Attention-based time series forecasting with learnable and interpretable basis. Advances in Neural Information Processing Systems, 71222--71241

2023
[21]

H.; Sinthong, P.; and Kalagnanam, J

Nie, Y.; Nguyen, N. H.; Sinthong, P.; and Kalagnanam, J. 2023. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers. In International Conference on Learning Representations

2023
[22]

Rasul, K.; Seward, C.; Schuster, I.; and Vollgraf, R. 2021. Autoregressive denoising diffusion models for multivariate probabilistic time series forecasting. In Proceedings of the International Conference on Machine Learning, 8857--8868

2021
[23]

Shen, L.; and Kwok, J. T. 2023. Non-autoregressive conditional diffusion models for time series prediction. In Proceedings of the International Conference on Machine Learning, 31016--31029

2023
[24]

A.; Maheswaranathan, N.; and Ganguli, S

Sohl - Dickstein, J.; Weiss, E. A.; Maheswaranathan, N.; and Ganguli, S. 2015. Deep Unsupervised Learning using Nonequilibrium Thermodynamics. In Proceedings of the International Conference on Machine Learning, 2256--2265

2015
[25]

Woo, G.; Liu, C.; Kumar, A.; Xiong, C.; Savarese, S.; and Sahoo, D. 2024. Unified Training of Universal Time Series Forecasting Transformers. In Proceedings of the International Conference on Machine Learning, 53140--53164

2024
[26]

Yang, Y.; Jin, M.; Wen, H.; Zhang, C.; Liang, Y.; Ma, L.; Wang, Y.; Liu, C.; Yang, B.; Xu, Z.; Bian, J.; Pan, S.; and Wen, Q. 2024. A Survey on Diffusion Models for Time Series and Spatio-Temporal Data. arXiv preprint, arXiv:2404.18886

2024
[27]

Yoon, J.; Jarrett, D.; and van der Schaar, M. 2019. Time-series Generative Adversarial Networks. In Advances in Neural Information Processing Systems, 5509--5519

2019
[28]

Yuan, X.; and Qiao, Y. 2024. Diffusion-TS: Interpretable Diffusion for General Time Series Generation. In International Conference on Learning Representations

2024

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

, " * write output.state after.block = add.period write newline

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

[2] [2]

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 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

[3] [3]

Coletta, A.; Gopalakrishnan, S.; Borrajo, D.; and Vyetrenko, S. 2023. On the Constrained Time-Series Generation Problem. In Advances in Neural Information Processing Systems, 61048--61059

work page 2023

[4] [4]

Coletta, A.; Prata, M.; Conti, M.; Mercanti, E.; Bartolini, N.; Moulin, A.; Vyetrenko, S.; and Balch, T. 2021. Towards realistic market simulations: a generative adversarial networks approach. In Proceedings of the ACM International Conference on AI in Finance, 1--9

work page 2021

[5] [5]

Das, A.; Kong, W.; Sen, R.; and Zhou, Y. 2024. A decoder-only foundation model for time-series forecasting. In Proceedings of the International Conference on Machine Learning, 10148--10167

work page 2024

[6] [6]

Desai, A.; Freeman, C.; Wang, Z.; and Beaver, I. 2021. TimeVAE: A Variational Auto-Encoder for Multivariate Time Series Generation. arXiv preprint, arXiv:2111.08095

arXiv 2021

[7] [7]

Fan, X.; Wu, Y.; Xu, C.; Huang, Y.; Liu, W.; and Bian, J. 2024. MG-TSD: Multi-Granularity Time Series Diffusion Models with Guided Learning Process. In International Conference on Learning Representations

work page 2024

[8] [8]

Gao, S.; Koker, T.; Queen, O.; Hartvigsen, T.; Tsiligkaridis, T.; and Zitnik, M. 2024. UniTS: Building a Unified Time Series Model. arXiv preprint, arXiv:2403.00131

arXiv 2024