Pith. sign in

REVIEW 3 major objections 6 minor 2 cited by

SynEHRgy: Synthesizing Mixed-Type Structured Electronic Health Records using Decoder-Only Transformers

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A small decoder-only transformer can generate synthetic electronic health records spanning demographics, ICD codes, and irregularly sampled time series.

desk verdict Per-type fidelity and a useful quantization ablation, but the paper's headline mixed-type claim is never actually tested on joint dependencies. read the letter →

arxiv 2411.13428 v1 pith:GZHGIILI submitted 2024-11-20 cs.LG cs.AI

Hojjat Karami , David Atienza , Anisoara Ionescu This is my paper · ORCID
classification cs.LGcs.AI
keywords syntheticelectronichealthrecordsdecoder-onlytransformertokenizationirregularlysampledtimeseriesICDcodesMIMIC-IIImembershipinferenceattackdataaugmentation
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

This paper tries to show that one small GPT-style decoder-only transformer can generate entire synthetic patient records that mix demographics, ICD diagnosis and procedure codes, and irregularly sampled vital-sign and lab time series across multiple hospital visits. The key trick is a tokenization scheme that turns every continuous value into one of a fixed set of bin tokens and recovers a number by sampling uniformly inside the bin. On the MIMIC-III critical care data, the generated records score competitively with, and on several n-gram and time-series fidelity metrics better than, specialized GAN and transformer baselines, and they support train-on-synthetic/test-on-real predictions nearly as well as real training data. If this holds, a single autoregressive language model can stand in for a collection of task-specific generative models for structured EHRs.

What carries the argument

The load-bearing mechanism is the tokenization strategy for mixed-type records. Every continuous value is uniformly quantized into equal-width bins, each bin getting its own token; time intervals between irregular measurements are similarly discretized; ICD codes and categorical variables get one token each; and special tokens (<s>, </s>, </covars>, </labels>, </ts>, </visit>, <PAD>) mark record structure. A small GPT-2-style decoder-only transformer is trained with the causal language-modeling objective to predict the next token, and generation proceeds autoregressively from <s> until </s>; numerical values are recovered by uniform sampling inside the predicted bin. This machinery lets one model handle discrete and continuous data in a single sequence while keeping token counts small.

What would settle it

Generate synthetic time series with the published pipeline and compare real and generated values within each bin, for example with a Kolmogorov-Smirnov test on heart-rate values inside a single bin; if within-bin values are systematically off, or if changing the number of bins from coarse to fine swings the reported fidelity and utility scores by a large margin, the central claim that binning preserves the distribution is false.

Watch

Extended reading notes

Core claim

The central claim is that discretizing structured EHR data into tokens and learning their next-token distribution lets a decoder-only transformer capture the joint distribution over covariates, ICD codes, and irregular time series well enough for practical synthetic-data uses. The authors demonstrate this on MIMIC-III with a small GPT-2-sized model, measuring fidelity through n-gram correlations for codes and precision/recall/density/coverage plus correlation-matrix error for time-series embeddings, utility through phenotype and mortality prediction with varying amounts of synthetic augmentation, and privacy through membership-inference distance statistics. Their reported results show the method ahead of the baselines on bigram and trigram code fidelity and on time-series precision, recall, and coverage, and closest to the real-data ceiling in train-on-synthetic/test-on-real utility. The paper frames the numerical-tokenization step as the load-bearing innovation: without it, the ablated text-based model performs markedly worse.

Load-bearing premise

The method assumes that chopping every continuous measurement into a fixed set of value bins and then drawing a random value inside the bin preserves the clinically meaningful distribution of the original measurements.

Editorial extensions

If this is right

  • A single decoder-only transformer can jointly model discrete codes and irregular continuous measurements, so mixed-type EHR generation does not require separate GAN or diffusion components per data type.
  • Irregularly sampled time series with informative missingness can be represented faithfully as token sequences, as shown by the co-occurrence and correlation-matrix results.
  • Synthetic records from this method can replace real training data for mortality and phenotype prediction with only a small AUROC drop, and can augment small training sets.
  • Because new variables only require adding tokens to the dictionary, the tokenization extends to new data sources without rearchitecting the model.
  • The approach inherits the context-length limit: very long patient histories may exceed the 1024-token window used here, a limitation the paper states in its conclusion.

Reading between the lines

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

  • The bin width and number of bins are not reported; if treated as a tunable hyperparameter, the same pipeline could plausibly be pushed toward higher fidelity with a finer grid or stronger privacy with a coarser grid, so the reported balance is not a fixed property of the method.
  • Because the model sees only bin identities, it cannot generate values outside the observed range or capture sub-bin structure; downstream tasks that need exact lab values, such as medication dosing rules, would require a different de-tokenization or a second-stage model.
  • The membership-inference evaluation uses distance distributions that can miss some memorization; a stronger attack targeting exact bin-token sequences might reveal more leakage than the reported near-chance AUROC.
  • The same tokenization recipe could transfer to non-medical mixed-type longitudinal records, such as sensor logs or administrative event streams, where irregular timing and categorical-numerical mixtures are common.
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 / 6 minor

Summary. The paper presents SynEHRgy, a method for generating synthetic longitudinal electronic health records with a decoder-only transformer. It tokenizes demographics, labels, ICD codes, irregularly sampled time series, and time intervals, trains a small GPT-2-style model on token sequences from MIMIC-III, and generates new patient records autoregressively. Evaluation covers per-type fidelity (ICD n-gram correlations; PRDC, correlation MSE, and missingness co-occurrence for time series), utility (TSTR and data-augmentation AUROC for phenotype and mortality prediction), and privacy (membership-inference distances). The central claim is that a single GPT-like decoder-only transformer with a tailored tokenization strategy can generate high-quality mixed-type structured EHR data across multiple visits.

Significance. If the empirical claims hold, SynEHRgy is a useful demonstration that a generic causal language model can match or beat specialized generative models for structured EHR data, especially irregularly sampled time series. Strengths include the use of a public benchmark, comparison against several relevant baselines, and evaluation along fidelity, utility, and privacy axes; the TSTR AUROC is close to the validation-split reference. However, the reported evidence is mostly per-data-type and does not directly verify the paper's central 'mixed-type' claim. The tokenization details are also not fully disclosed, which limits reproducibility and makes the main mechanism difficult to audit.

major comments (3)
  1. [Section 4.2, Tables 1-4] The abstract and introduction claim generation of mixed-type structured EHRs, but the evaluation never checks dependencies across data types. Section 4.2 evaluates ICD fidelity with n-gram correlations and time-series fidelity with PRDC/MSE_corr separately, and the utility tasks in Tables 3 and 4 use time-series embeddings only and do not involve ICD codes. A model could match all reported per-type statistics while generating clinically inconsistent joint records (e.g., normal vitals paired with septic-shock codes). This is a missing verification of the central claim. Please add a joint fidelity test, such as comparing the conditional distribution of time-series statistics given ICD codes or phenotypes between real and synthetic data, or a discriminative test on concatenated representations.
  2. [Sections 3.2 and 3.4] The method's numerical tokenization relies on uniform quantization, but the paper reports neither the number nor the width of bins per variable. De-tokenization uses uniform sampling inside each bin, so the generated continuous values cannot reproduce the within-bin distribution unless the bins are very narrow. This choice is an unauditable free parameter that affects every fidelity, utility, and privacy result. Please report the full binning configuration and add a sensitivity analysis (e.g., coarse versus fine grids) to show that the method is robust to this choice.
  3. [Section 5.1 and NeurIPS checklist item 7] The main empirical claims are not accompanied by error bars in Tables 1, 3, 4, and 5, and the standard deviations in Table 2 are not explained (over random seeds, train splits, or bootstrap samples). The checklist explicitly acknowledges that k-fold or multi-seed evaluation was not performed. Without uncertainty estimates, the reported differences that are small, such as utility AUROCs and MIA metrics, cannot be distinguished from noise. Please provide confidence intervals or multiple-run variability for the key comparisons, at least for the central fidelity and TSTR claims.
minor comments (6)
  1. [Section 4.4] The phrase 'context length of size 4098' appears to be a typo; presumably 4096 was intended.
  2. [Table 2] Please state explicitly what the parenthetical values are (standard deviation over which source of randomness, such as seeds or bootstrap samples).
  3. [Figure 3] The missingness co-occurrence is only shown visually; a numerical summary, such as the MSE between the co-occurrence matrices of real and synthetic data, would make the claim quantitative.
  4. [Related Work] CEHR-GPT is described as a GPT-based longitudinal EHR generator; consider including it as a baseline or explaining why a comparison is omitted.
  5. [Checklist item 12] The license question is answered as NA although the paper uses MIMIC-III; the dataset's license and terms of use should be stated.
  6. [Section 3.4] The statement that uniform sampling 'ensures that the generated values are representative' is too strong, since uniform sampling within a bin ignores the distribution of values inside that bin.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the evaluation is empirical and self-contained against external benchmarks.

full rationale

The paper makes no formal derivation or first-principles prediction whose output is equivalent to its input. The central claim is empirical: a decoder-only transformer with the proposed tokenization generates synthetic EHRs whose fidelity, utility, and privacy are measured against the held-out MIMIC-III test split and against external baselines such as HALO, PromptEHR, RTSGAN, and TimEHR. The tokenization and de-tokenization procedures (uniform quantization into bins, then uniform sampling within the bin) are modeling choices, not fitted quantities that are later renamed as predictions. The evaluation metrics are applied symmetrically to the proposed method and to baselines, and the test split is used as a real-data reference rather than as a fitted target. The only self-citation is the authors' own TimEHR baseline, which appears in related work and as a comparison method; it is not load-bearing for the method's validity or for any theoretical uniqueness claim. Accordingly, no circular step can be exhibited with a specific reduction, and the appropriate finding is a low score reflecting only a minor, non-load-bearing self-citation.

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

No formal derivation is present, so the ledger records hand-chosen hyperparameters and domain assumptions. The main unquantified choice is the binning grid for numerical values and time intervals. No new physical or conceptual entities are introduced.

free parameters (5)
  • Uniform quantization bin grid per numerical variable = Not reported
    Section 3.2 discretizes numerical values into equal-sized bins; the example '<HR_0> represents heart rate values in the range of 55-66' implies a fixed grid, but the full grid is not specified. All numerical fidelity depends on this choice.
  • Time interval quantization bin grid = Not reported
    Section 3.2 discretizes intervals between consecutive measurements into bins; the bin sizes are not specified, and this controls how temporal irregularity is represented.
  • Context length = 1024 tokens
    Section 4.4: chosen as a balance between quality and training time; it covers 51 percent of the original data, so long patient histories are truncated.
  • Generation sampling temperature and top-k = temperature 0.7, top-k 50
    Section 4.4: these generation-time hyperparameters control the diversity and fidelity of the sampled records.
  • Model and training hyperparameters = 4 layers, 4 heads, 384 dims, lr 3e-4, 20 epochs, batch 128
    Section 4.4: chosen with minimal tuning; they shape the learned distribution and are not justified by a systematic search.
assumptions (5)
  • domain assumption Autoregressive next-token prediction over the tokenized stream approximates the joint distribution of covariates, codes, and irregular time series.
    Invoked throughout Section 3.3; if the causal language modeling objective cannot capture cross-type dependencies, the generated records will not be realistic.
  • domain assumption Uniform quantization followed by uniform within-bin sampling preserves the distribution of continuous clinical values.
    Used in Sections 3.2 and 3.4; this is the central numerical reconstruction assumption.
  • domain assumption The MIMIC-III preprocessing pipeline of Harutyunyan et al. defines the target distribution.
    Section 4.1 uses this pipeline to extract covariates, codes, and time series; all experiments inherit its choices.
  • domain assumption The distance-based membership inference attack from EHR-Safe is a valid privacy test.
    Section 4.2 and Table 5 use this attack; the privacy conclusion depends on its adequacy.
  • domain assumption PRDC metrics and n-gram correlations are sufficient fidelity measures.
    Section 4.2 defines fidelity via these metrics; no external clinical validation is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SynEHRgy: Synthesizing Mixed-Type Structured Electronic Health Records using Decoder-Only Transformers." pith.science (2026). https://pith.science/paper/GZHGIILI

@misc{pith2026241113428,
  author       = {Pith},
  title        = {Pith review of: SynEHRgy: Synthesizing Mixed-Type Structured Electronic Health Records using Decoder-Only Transformers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GZHGIILI}},
  note         = {Machine review of arXiv:2411.13428}
}
read the original abstract

Generating synthetic Electronic Health Records (EHRs) offers significant potential for data augmentation, privacy-preserving data sharing, and improving machine learning model training. We propose a novel tokenization strategy tailored for structured EHR data, which encompasses diverse data types such as covariates, ICD codes, and irregularly sampled time series. Using a GPT-like decoder-only transformer model, we demonstrate the generation of high-quality synthetic EHRs. Our approach is evaluated using the MIMIC-III dataset, and we benchmark the fidelity, utility, and privacy of the generated data against state-of-the-art models.

Figures

Figures reproduced from arXiv: 2411.13428 by the authors.

Figure 1
Figure 1. Left: Example of patient’s data. Right: Tokenized patient’s data [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Confusion matrices of correlation levels for time series data [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Co-occurrence of measurement for time series data [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Privacy-Preserving Tabular Synthetic Data Generation Using TabularARGN

    cs.LG 2025-08 unverdicted novelty 5.0 of 10

    TabularARGN is a discretization-based auto-regressive network claimed to generate high-fidelity, privacy-robust synthetic tabular data, competitive with diffusion and GAN baselines.

  2. TabularARGN: A Flexible and Efficient Auto-Regressive Framework for Generating High-Fidelity Synthetic Data

    cs.LG 2025-01 conditional novelty 5.0 of 10

    TabularARGN is a lightweight any-order auto-regressive model that matches or beats state-of-the-art tabular synthetic data quality while training up to orders of magnitude faster, including for sequential two-table datasets.

Reference graph

Works this paper leans on

64 extracted references · 31 canonical work pages · cited by 2 Pith papers

  1. [1]

    The Future of Digital Health with Federated Learning

    Nicola Rieke et al. “The Future of Digital Health with Federated Learning”. In: npj Digital Medicine 3.1 (Sept. 2020), p. 119. ISSN : 2398-6352. DOI: 10.1038/s41746-020-00323-1 . arXiv: 2003.08119 [cs]

    work page Pith review arXiv 2020

  2. [2]

    Privacy-Preserving Artificial Intelligence in Healthcare: Techniques and Applications

    Nazish Khalid et al. “Privacy-Preserving Artificial Intelligence in Healthcare: Techniques and Applications”. In: Computers in Biology and Medicine 158 (May 2023), p. 106848. ISSN : 0010-4825. DOI: 10.1016/j.compbiomed.2023.106848

    arXiv 2023

  3. [3]

    A Review of Challenges and Opportunities in Machine Learning for Health

    Marzyeh Ghassemi et al. “A Review of Challenges and Opportunities in Machine Learning for Health”. In: AMIA Summits on Translational Science Proceedings 2020 (May 2020), pp. 191–

    work page 2020

  4. [4]

    Informative Missingness: What Can We Learn from Patterns in Miss- ing Laboratory Data in the Electronic Health Record?

    Amelia L. M. Tan et al. “Informative Missingness: What Can We Learn from Patterns in Miss- ing Laboratory Data in the Electronic Health Record?” In: Journal of Biomedical Informatics 139 (Mar. 2023), p. 104306. ISSN : 1532-0464. DOI: 10.1016/j.jbi.2023.104306

    arXiv 2023

  5. [5]

    Generating Multi-label Discrete Patient Records Using Generative Adver- sarial Networks

    Edward Choi et al. “Generating Multi-label Discrete Patient Records Using Generative Adver- sarial Networks”. In: arXiv: Learning (Mar. 2017)

    work page 2017

  6. [6]

    Fox.CorGAN: Correlation-Capturing Convolutional Generative Adversarial Networks for Generating Synthetic Healthcare Records

    Amirsina Torfi and Edward A. Fox.CorGAN: Correlation-Capturing Convolutional Generative Adversarial Networks for Generating Synthetic Healthcare Records . Mar. 2020. DOI: 10 . 48550/arXiv.2001.09346. arXiv: 2001.09346 [cs, stat]

    arXiv doi:10.48550/arxiv.2001.09346 2020

  7. [7]

    EVA: Generating Longitudinal Electronic Health Records Using Conditional Variational Autoencoders

    Siddharth Biswal et al. EVA: Generating Longitudinal Electronic Health Records Using Conditional Variational Autoencoders. Dec. 2020. DOI: 10.48550/arXiv.2012.10020 . arXiv: 2012.10020 [cs]

    work page Pith review arXiv doi:10.48550/arxiv.2012.10020 2020

  8. [8]

    PromptEHR: Conditional Electronic Healthcare Records Generation with Prompt Learning

    Zifeng Wang and Jimeng Sun. PromptEHR: Conditional Electronic Healthcare Records Generation with Prompt Learning. Oct. 2022. DOI: 10.48550/arXiv.2211.01761. arXiv: 2211.01761 [cs]

    work page Pith review arXiv doi:10.48550/arxiv.2211.01761 2022

Show all 64 references
  1. [9]

    Towards Generating Real-World Time Series Data

    Hengzhi Pei et al. “Towards Generating Real-World Time Series Data”. In: (Dec. 2021).DOI: 10.1109/ICDM51629.2021.00058

    2021

  2. [10]

    Fast and Reliable Generation of EHR Time Series via Diffusion Models

    Muhang Tian et al. Fast and Reliable Generation of EHR Time Series via Diffusion Models. Oct. 2023. DOI: 10.48550/arXiv.2310.15290. arXiv: 2310.15290 [cs]

    2023 doi arXiv

  3. [11]

    TimEHR: Image-based Time Series Generation for Electronic Health Records

    Hojjat Karami et al. TimEHR: Image-based Time Series Generation for Electronic Health Records. Feb. 2024. DOI: 10.48550/arXiv.2402.06318. arXiv: 2402.06318 [cs]

    2024 doi arXiv

  4. [12]

    Synthesize High-Dimensional Longitudinal Electronic Health Records via Hierarchical Autoregressive Language Model

    Brandon Theodorou, Cao Xiao, and Jimeng Sun. “Synthesize High-Dimensional Longitudinal Electronic Health Records via Hierarchical Autoregressive Language Model”. In: Nature Communications 14.1 (Aug. 2023), p. 5305. ISSN : 2041-1723. DOI: 10.1038/s41467-023- 41093-0

    2023 doi

  5. [13]

    BEHRT: Transformer for Electronic Health Records

    Yikuan Li et al. BEHRT: Transformer for Electronic Health Records. July 2019. DOI: 10. 48550/arXiv.1907.09538. arXiv: 1907.09538 [cs, stat]

    2019 doi arXiv

  6. [14]

    ClinicalBERT: Modeling Clinical Notes and Predicting Hospital Readmission

    Kexin Huang, Jaan Altosaar, and Rajesh Ranganath. ClinicalBERT: Modeling Clinical Notes and Predicting Hospital Readmission . Nov. 2020. DOI: 10.48550/arXiv.1904.05342 . arXiv: 1904.05342 [cs]

    2020 doi arXiv

  7. [15]

    Towards Expert-Level Medical Question Answering with Large Language Models

    Karan Singhal et al. Towards Expert-Level Medical Question Answering with Large Language Models. May 2023. DOI: 10.48550/arXiv.2305.09617. arXiv: 2305.09617 [cs]. 8

    2023 doi arXiv

  8. [16]

    Zeming Chen et al.MEDITRON-70B: Scaling Medical Pretraining for Large Language Models. Nov. 2023. DOI: 10.48550/arXiv.2311.16079. arXiv: 2311.16079 [cs]

    2023 doi arXiv

  9. [17]

    MIMIC-III, a Freely Accessible Critical Care Database

    Alistair E. W. Johnson et al. “MIMIC-III, a Freely Accessible Critical Care Database”. In: Scientific Data 3.1 (May 2016), p. 160035. ISSN : 2052-4463. DOI: 10.1038/sdata.2016.35

    2016 doi

  10. [18]

    Synthesizing Electronic Health Records Using Improved Genera- tive Adversarial Networks

    Mrinal Kanti Baowaly et al. “Synthesizing Electronic Health Records Using Improved Genera- tive Adversarial Networks”. In: Journal of the American Medical Informatics Association 26.3 (Mar. 2019), pp. 228–241. ISSN : 1067-5027, 1527-974X. DOI: 10.1093/jamia/ocy142

    2019 doi

  11. [19]

    SynTEG: A Framework for Temporal Structured Electronic Health Data Simulation

    Ziqi Zhang et al. “SynTEG: A Framework for Temporal Structured Electronic Health Data Simulation”. In: Journal of the American Medical Informatics Association 28.3 (Mar. 2021), pp. 596–604. ISSN : 1527-974X. DOI: 10.1093/jamia/ocaa262

    2021 doi

  12. [20]

    Multi-Label Clinical Time-Series Generation via Conditional GAN

    Chang Lu et al. “Multi-Label Clinical Time-Series Generation via Conditional GAN”. In: IEEE Transactions on Knowledge and Data Engineering (2023), pp. 1–13. ISSN : 1558-2191. DOI: 10.1109/TKDE.2023.3310909

    2023

  13. [21]

    Synthetic Electronic Health Records Generated with Variational Graph Autoencoders

    Giannis Nikolentzos et al. “Synthetic Electronic Health Records Generated with Variational Graph Autoencoders”. In: npj Digital Medicine 6.1 (Apr. 2023), pp. 1–12. ISSN : 2398-6352. DOI: 10.1038/s41746-023-00822-x

    2023 doi

  14. [22]

    MedDiff: Generating Electronic Health Records Using Accelerated Denoising Diffusion Model

    Huan He et al. MedDiff: Generating Electronic Health Records Using Accelerated Denoising Diffusion Model. Feb. 2023. DOI: 10 . 48550 / arXiv . 2302 . 04355. arXiv: 2302 . 04355 [cs]

    2023

  15. [23]

    Synthesizing Mixed-type Electronic Health Records Using Diffusion Models

    Taha Ceritli et al. Synthesizing Mixed-type Electronic Health Records Using Diffusion Models. Aug. 2023. DOI: 10.48550/arXiv.2302.14679. arXiv: 2302.14679 [cs]

    2023 doi arXiv

  16. [24]

    Hongyi Yuan, Songchi Zhou, and Sheng Yu.EHRDiff: Exploring Realistic EHR Synthesis with Diffusion Models. Mar. 2024. DOI: 10.48550/arXiv.2303.05656 . arXiv: 2303.05656 [cs]

    2024 doi arXiv

  17. [25]

    Synthesizing Multimodal Electronic Health Records via Predictive Diffusion Models

    Yuan Zhong et al. Synthesizing Multimodal Electronic Health Records via Predictive Diffusion Models. June 2024. DOI: 10.48550/arXiv.2406.13942. arXiv: 2406.13942 [cs]

    2024 doi arXiv

  18. [26]

    CEHR-GPT: Generating Electronic Health Records with Chronological Patient Timelines

    Chao Pang et al. CEHR-GPT: Generating Electronic Health Records with Chronological Patient Timelines. May 2024. DOI: 10.48550/arXiv.2402.04400 . arXiv: 2402.04400 [cs]

    2024 doi arXiv

  19. [27]

    Real-Valued (Medical) Time Series Generation with Recurrent Conditional GANs

    Cristóbal Esteban, Stephanie L. Hyland, and Gunnar Rätsch. “Real-Valued (Medical) Time Series Generation with Recurrent Conditional GANs”. In: arXiv: Machine Learning (June 2017)

    2017

  20. [28]

    Time-Series Generative Adversarial Networks

    Jinsung Yoon, Daniel Jarrett, and Mihaela van der Schaar. “Time-Series Generative Adversarial Networks”. In: (Sept. 2019)

    2019

  21. [29]

    Using GANs for Sharing Networked Time Series Data: Challenges, Initial Promise, and Open Questions

    Zinan Lin et al. “Using GANs for Sharing Networked Time Series Data: Challenges, Initial Promise, and Open Questions”. In: Proceedings of the ACM Internet Measurement Conference. Virtual Event USA: ACM, Oct. 2020, pp. 464–483. ISBN : 978-1-4503-8138-3. DOI: 10.1145/ 3419394.3423643

    2020

  22. [30]

    Generating Synthetic Mixed-Type Longitudinal Electronic Health Records for Artificial Intelligent Applications

    Jin Li et al. “Generating Synthetic Mixed-Type Longitudinal Electronic Health Records for Artificial Intelligent Applications”. In: npj Digital Medicine 6.1 (May 2023), pp. 1–18. ISSN : 2398-6352. DOI: 10.1038/s41746-023-00834-7

    2023 doi

  23. [31]

    EHR-Safe: Generating High-Fidelity and Privacy-Preserving Synthetic Electronic Health Records

    Jinsung Yoon et al. “EHR-Safe: Generating High-Fidelity and Privacy-Preserving Synthetic Electronic Health Records”. In: npj Digital Medicine 6.1 (Aug. 2023), pp. 1–11. ISSN : 2398-

    2023

  24. [32]

    Diffusion-TS: Interpretable Diffusion for General Time Series Generation

    Xinyu Yuan and Yan Qiao. “Diffusion-TS: Interpretable Diffusion for General Time Series Generation”. In: The Twelfth International Conference on Learning Representations . Oct. 2023

    2023

  25. [33]

    TimeLDM: Latent Diffusion Model for Unconditional Time Series Generation

    Jian Qian et al. TimeLDM: Latent Diffusion Model for Unconditional Time Series Generation. July 2024. arXiv: 2407.04211 [cs]

    2024 arXiv

  26. [34]

    TimeAutoDiff: Combining Autoencoder and Diffusion Model for Time Series Tabular Data Synthesizing

    Namjoon Suh et al. TimeAutoDiff: Combining Autoencoder and Diffusion Model for Time Series Tabular Data Synthesizing. https://arxiv.org/abs/2406.16028v2

  27. [35]

    TS-Diffusion: Generating Highly Complex Time Series with Diffusion Models

    Yangming Li. TS-Diffusion: Generating Highly Complex Time Series with Diffusion Models. Nov. 2023. DOI: 10.48550/arXiv.2311.03303. arXiv: 2311.03303 [cs]

    2023 doi arXiv

  28. [36]

    Tuan Dinh et al.LIFT: Language-Interfaced Fine-Tuning for Non-Language Machine Learning Tasks. Oct. 2022. DOI: 10.48550/arXiv.2206.06565. arXiv: 2206.06565 [cs]. 9

    2022 doi arXiv

  29. [37]

    Language Models Are Realistic Tabular Data Generators

    Vadim Borisov et al. Language Models Are Realistic Tabular Data Generators. Apr. 2023. DOI: 10.48550/arXiv.2210.06280. arXiv: 2210.06280 [cs]

    2023 doi arXiv

  30. [38]

    Leveraging VQ-V AE Tokenization for Autoregressive Modeling of Medical Time Series

    Yoonhyung Lee, Younhyung Chae, and Kyomin Jung. “Leveraging VQ-V AE Tokenization for Autoregressive Modeling of Medical Time Series”. In: Artificial Intelligence in Medicine 154 (Aug. 2024), p. 102925. ISSN : 0933-3657. DOI: 10.1016/j.artmed.2024.102925

    2024

  31. [39]

    Chronos: Learning the Language of Time Series

    Abdul Fatir Ansari et al. Chronos: Learning the Language of Time Series. Mar. 2024. arXiv: 2403.07815 [cs]

    2024 arXiv

  32. [40]

    Synthetic Time Series Generation for Decision Intelli- gence Using Large Language Models

    Alexandru Grigoras, and Florin Leon. “Synthetic Time Series Generation for Decision Intelli- gence Using Large Language Models”. In: Mathematics 12.16 (Aug. 2024), p. 2494. ISSN : 2227-7390. DOI: 10.3390/math12162494

    2024 doi

  33. [41]

    Language Models Are Unsupervised Multitask Learners

    Alec Radford et al. “Language Models Are Unsupervised Multitask Learners”. In: OpenAI blog 1.8 (2019), p. 9

    2019

  34. [42]

    Multitask Learning and Benchmarking with Clinical Time Series Data

    Hrayr Harutyunyan et al. “Multitask Learning and Benchmarking with Clinical Time Series Data”. In: Scientific Data 6.1 (June 2019), p. 96. ISSN : 2052-4463. DOI: 10.1038/s41597- 019-0103-9. arXiv: 1703.07771 [cs, stat]

    2019 arXiv

  35. [43]

    Reliable Fidelity and Diversity Metrics for Generative Models

    Muhammad Ferjad Naeem et al. Reliable Fidelity and Diversity Metrics for Generative Models. June 2020. DOI: 10.48550/arXiv.2002.09797. arXiv: 2002.09797 [cs, stat]

    2020 doi arXiv

  36. [44]

    LightGBM: A Highly Efficient Gradient Boosting Decision Tree

    Guolin Ke et al. “LightGBM: A Highly Efficient Gradient Boosting Decision Tree”. In: Advances in Neural Information Processing Systems. V ol. 30. Curran Associates, Inc., 2017

    2017

  37. [45]

    Unified Training of Universal Time Series Forecasting Transformers

    Gerald Woo et al. Unified Training of Universal Time Series Forecasting Transformers. Feb

  38. [46]

    June 2024

    Yong Liu et al.Timer: Generative Pre-trained Transformers Are Large Time Series Models. June 2024. DOI: 10.48550/arXiv.2402.02368. arXiv: 2402.02368 [cs, stat]

    2024 doi arXiv

  39. [47]

    [Yes] " is generally preferable to

    Karan Singhal et al. Large Language Models Encode Clinical Knowledge. Dec. 2022. DOI: 10.48550/arXiv.2212.13138. arXiv: 2212.13138 [cs]. 10 NeurIPS Paper Checklist The checklist is designed to encourage best practices for responsible machine learning research, addressing issue...

    2022 doi arXiv

  40. [50]

    We have deeply evaluated the quality of synthetic data in terms of utility, fidelity and privacy and compared it with the state-of-the-art models

    Claims Question: Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? Answer: [Yes] Justification: Our work is mostly empirical rather than theoretical. We have deeply evaluated the quality of synthetic data in terms ...

  41. [51]

    Limitations

    Limitations Question: Does the paper discuss the limitations of the work performed by the authors? Answer: [Yes] 11 Justification: In the conclusion section, we discussed the limitations of our work such as the tokenization strategy might not be efficient for very large sequen...

  42. [52]

    Guidelines: • The answer NA means that the paper does not include theoretical results

    Theory Assumptions and Proofs Question: For each theoretical result, does the paper provide the full set of assumptions and a complete (and correct) proof? Answer: [NA] Justification: Our work is mostly empirical and does not include theoretical results. Guidelines: • The answ...

  43. [53]

    The model architecture and training details are provided in the main paper

    Experimental Result Reproducibility Question: Does the paper fully disclose all the information needed to reproduce the main ex- perimental results of the paper to the extent that it affects the main claims and/or conclusions of the paper (regardless of whether the code and da...

  44. [54]

    We will release the code and preprocessing pipeline upon acceptance

    Open access to data and code Question: Does the paper provide open access to the data and code, with sufficient instruc- tions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: The dataset (MIMIC-III) is ...

  45. [55]

    The labels for downstream tasks are also selected using the same pipeline

    Experimental Setting/Details Question: Does the paper specify all the training and test details (e.g., data splits, hyper- parameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? Answer: [Yes] Justification: We have used a famous preproc...

  46. [56]

    Experiment Statistical Significance Question: Does the paper report error bars suitably and correctly defined or other appropriate information about the statistical significance of the experiments? Answer: [No] Justification: As our training split is large enough ( 30k), we ha...

  47. [57]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiments Compute Resources 14 Question: For each experiment, does the paper provide sufficient information on the com- puter resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [Yes] Justification: This information is ...

  48. [58]

    We have tried our best to preserve the anonymity

    Code Of Ethics Question: Does the research conducted in the paper conform, in every respect, with the NeurIPS Code of Ethics https://neurips.cc/public/EthicsGuidelines? Answer: [Yes] Justification: Our dataset is publicly available and we have used a widely used model architec...

  49. [59]

    Broader Impacts Question: Does the paper discuss both potential positive societal impacts and negative societal impacts of the work performed? Answer: [Yes] Justification: The development of SynEHRgy, while offering significant advances in the generation of synthetic EHR data,...

  50. [60]

    Safeguards Question: Does the paper describe safeguards that have been put in place for responsible release of data or models that have a high risk for misuse (e.g., pretrained language models, image generators, or scraped datasets)? Answer: [NA] Justification: [NA] Guidelines...

  51. [61]

    • The authors should cite the original paper that produced the code package or dataset

    Licenses for existing assets Question: Are the creators or original owners of assets (e.g., code, data, models), used in the paper, properly credited and are the license and terms of use explicitly mentioned and properly respected? Answer: [NA] Justification: [NA] Guidelines: ...

  52. [62]

    • Researchers should communicate the details of the dataset/code/model as part of their submissions via structured templates

    New Assets Question: Are new assets introduced in the paper well documented and is the documentation provided alongside the assets? Answer: [NA] Justification: [NA] Guidelines: • The answer NA means that the paper does not release new assets. • Researchers should communicate t...

  53. [63]

    Crowdsourcing and Research with Human Subjects Question: For crowdsourcing experiments and research with human subjects, does the paper include the full text of instructions given to participants and screenshots, if applicable, as well as details about compensation (if any)? A...

  54. [64]

    • Depending on the country in which research is conducted, IRB approval (or equivalent) may be required for any human subjects research

    Institutional Review Board (IRB) Approvals or Equivalent for Research with Human Subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals...

  55. [2024]

    arXiv: 2402.02592 [cs]

    DOI: 10.48550/arXiv.2402.02592. arXiv: 2402.02592 [cs]

    doi arXiv

  56. [6352]

    DOI: 10.1038/s41746-023-00888-7

    doi

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.