Pith. sign in

REVIEW 4 major objections 5 minor 42 references

Term2Note: Synthesising Differentially Private Clinical Notes from Medical Terms

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Term2Note synthesises long clinical notes under formal differential privacy by separating medical terms from note text, and reports that the synthetic notes preserve structure and downstream ICD-coding utility nearly as well as real notes.

desk verdict Solid engineering, thorough evaluation, but the privacy proof is missing—this is a conditional accept at best. read the letter →

arxiv 2509.10882 v1 pith:CGJGPCD4 submitted 2025-09-13 cs.CL

Yuping Wu , Viktor Schlegel , Warren Del-Pinto , Srinivasan Nandakumar , Iqra Zahid , Yidan Sun , Usama Farghaly Omar , Amirah Jasmine
show 5 more authors
Arun-Kumar Kaliya-Perumal Chun Shen Tham Gabriel Connors Anil A Bharath Goran Nenadic
This is my paper · ORCID
classification cs.CL
keywords differentialprivacysyntheticclinicalnotesmedicaltermextractionICDcodingDP-SGDnotegenerationprivacy-preservingdatasharingMIMIC
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

Term2Note tackles the problem of sharing sensitive clinical notes by generating synthetic discharge notes under formal differential privacy, a guarantee that no single patient's data can change the output distribution beyond a controlled bound. The method splits each note into sections, extracts medical terms per section, privatises those terms with a DP mechanism, and trains a note generator under DP to write section text conditioned on the privatised terms. A quality maximiser selects the most fluent candidate note as post-processing, so the privacy guarantee is unchanged. The paper reports that the synthetic notes closely match real notes in length distribution, term overlap, and semantic alignment, and that ICD-code classifiers trained on them perform close to classifiers trained on real notes.

What carries the argument

The load-bearing object is a section-wise, term-conditioned generation pipeline. SECSPLIT standardises a clinical note into at most six sections; TERMEXT extracts SNOMED CT terms per section; the terms are embedded and perturbed by DPRP*, an adaptation of the random-projection DP mechanism that adds noise to the embeddings and their covariance matrix, performs SVD, and reconstructs from noisy singular vectors; a GPT-2 term generator produces synthetic terms; and a DP-trained Llama-3.2-1B or Gemma-3-1B note generator writes each section conditioned on the synthetic terms and previously generated sections, trained with FastDP-based DP-SGD. A perplexity-based quality maximiser selects the lowest-perplexity candidate note. The privacy proof combines the post-processing property and parallel composition over the disjoint term-generation and note-generation components.

What would settle it

Run the DPRP* algorithm on two neighbouring term lists that differ in exactly one medical term, embed both with the clinical embedder, and compare the resulting output distributions; if any measurable output set has a probability ratio exceeding $e^{\epsilon}$ (plus $\delta$) at the claimed parameters, the privacy claim is refuted. A simpler check is to compute the global sensitivity of the embedding function and verify whether the noise scale in Algorithm 2 is large enough; the paper reports no such calculation.

Watch

Extended reading notes

Core claim

The central claim is that long-form clinical note synthesis under DP becomes tractable when content and form are structurally separated: privatise the clinically salient terms that carry patient-specific information, then let a DP-trained language model generate section-wise free text conditioned on those terms. When both terms and notes are privatised, the overall privacy loss is bounded by $(\max(\epsilon_n, \epsilon_t), \max(\delta_n, \delta_t))$ via parallel composition and post-processing; when only notes are privatised, the loss is $(\epsilon_n, \delta_n)$. Empirically, Term2Note achieves the lowest KL divergence in length and term distributions and the highest MAUVE scores among DP baselines, and its ICD-coding utility remains close to real-data performance even at a strict budget of $\epsilon = 2$.

Load-bearing premise

The term-privatisation step DPRP* is assumed to be differentially private when applied to term embeddings, but the paper supplies no proof, no sensitivity bound for the embedding vectors, and no noise calibration; if that step is not DP, the formal privacy guarantee for the entire pipeline fails.

Editorial extensions

If this is right

  • If the central claim holds, hospitals and research consortia could share synthetic discharge notes with formal DP guarantees instead of raw notes, enabling multi-site machine learning without direct transfer of patient data.
  • The term-level privacy budget gives a tunable control: the same DP-trained note generator can be paired with different $\epsilon_t$ values to trade clinical content fidelity against privacy strength.
  • ICD-code classifiers trained on synthetic notes reach performance close to those trained on real notes, suggesting that other structured clinical NLP tasks may also transfer from such synthetic corpora.
  • Because the quality maximiser is a post-processing step, any selection or filtering rule applied to candidate notes preserves the privacy guarantee, making inference-time quality control a free privacy-safe lever.

Reading between the lines

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

  • Editorial inference: the formal privacy statement for the term path depends entirely on DPRP* being differentially private for the clinical embedder's outputs, which the paper does not prove; establishing a sensitivity bound for that embedding function would make the full-pipeline guarantee unconditional.
  • Editorial inference: the section-wise separation of content and form could transfer to other structured clinical documents such as radiology reports, operation notes, or discharge summaries from other institutions, since the mechanism is tied to section titles and SNOMED CT terms rather than to MIMIC formatting.
  • Editorial inference: the physicians' qualitative observations about medication misclassification and internal inconsistency suggest that safety-critical use of such synthetic notes would require a clinical fact-checking stage before deployment; this is an extension beyond the paper's claims.
  • Editorial inference: the utility evaluation collapses ICD codes into 20 broad groups, so a finer-grained ICD coding benchmark would be a stricter test of whether the synthetic fidelity translates to clinically specific prediction accuracy.
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.

Referee Report

4 major / 5 minor

Summary. The paper presents Term2Note, a framework for differentially private synthesis of full-length clinical notes. The method splits a source note into sections, extracts SNOMED-CT terms, privatizes the term embeddings with an adapted DPRP* mechanism, generates synthetic terms from the private embeddings, and then generates each note section with a language model fine-tuned under DP (FastDP). A DP quality maximiser selects among several candidate notes using an external perplexity model. Experiments on MIMIC-III/MIMIC-IV compare Term2Note against AUG-PE and FastDP baselines across fidelity, downstream ICD-coding utility, and human preference, reporting consistent improvements. The paper claims an overall (max(ε_n, ε_t), max(δ_n, δ_t))-DP guarantee for the full pipeline.

Significance. If the formal privacy claim were established, Term2Note would be a meaningful advance: it is, to my knowledge, the first attempt at DP synthesis of full-length clinical notes, and the empirical evaluation is broad, covering multiple privacy budgets, two base models, downstream task utility, and human clinical evaluation. The method is also refreshingly explicit about separating content (terms) from form (sections), and the reported fidelity improvements over strong baselines are substantively interesting. The supplied manuscript is transparent about many experimental details, including hyperparameters and supplementary ablations, which aids reproducibility. However, the central formal claim — the (ε, δ)-DP guarantee — depends on a privacy mechanism (DPRP*) whose DP property is asserted rather than proved, and whose pseudocode is not well-defined as written. Because this underpins the main contribution, the paper in its current form does not yet deliver the promised formal guarantee.

major comments (4)
  1. [§3.3, §3.6, Appendix A, Algorithm 2] The DP guarantee of DPRP* is asserted, not proven. The paper never gives a proof that Algorithm 2 is (ε_t, δ_t)-DP when applied to clinical term embeddings, nor does it state a sensitivity bound for the embedding vectors or for the covariance query E^T E. The embeddings come from MedEmbed-large and are not clipped or norm-bounded, so the L2 sensitivity of E^T E is not finite; consequently the Gaussian noise scales σ_1 and σ_2 in Algorithm 2 cannot be calibrated from (ε, δ) alone. Since Eq. (6) and the overall privacy claim both rest on the assumption that the TERMGEN component is (ε_t, δ_t)-DP, this missing analysis is load-bearing. The authors must supply a rigorous privacy proof (or, if the mechanism is not DP, revise the claim) and specify the sensitivity/noise calibration explicitly.
  2. [Appendix A, Algorithm 2, line 7] Line 7 of Algorithm 2, 'E_DP = E' V'_k^T + V'_k^T', is not a well-formed matrix expression as written: E' is n×d and V'_k is d×k, so E'V'_k^T is n×k, while V'_k^T is k×d; the two matrices cannot be added. The parenthetical comment that '+' refers to the Moore-Penrose pseudoinverse does not resolve the dimensional mismatch (the pseudoinverse of V'_k^T would be k×d, not n×k). As published, the algorithm is not reproducible, and the intended reconstruction operation must be clarified and stated in unambiguous notation.
  3. [§3.6, Appendix B] The composition argument does not correctly account for the unit of privacy. The paper treats the sections of a single note as if they were independent records when it scales the term-generation budget by 1/m, but the neighbouring relation in DP is defined at the level of the patient note. DPRP* is applied to the entire matrix of section embeddings of one note; replacing one note changes all m rows of E, so the sensitivity of the covariance query E^T E is not the sensitivity of a single row. The proof in Appendix B also assumes that M2 = 'TERMGEN processing on D_test' is (ε_t, δ_t)-DP without establishing this. A correct analysis must specify the neighbouring relation, derive the sensitivity of the full-note query, and show how the per-section budget split follows from that sensitivity.
  4. [§4, Table 1 and Table 7] The downstream utility evaluation is based on only 204 test notes with 80:20 five-fold cross-validation, and the reported standard deviations are large. For example, at ε=8, Term2Note with ε_t=∞ has macro F1 21.28±1.28 while FastDP has 16.40±4.01; at the same privacy cost, Term2Note with full privacy (ε_t=8) has macro F1 26.50±3.95. Several pairwise differences appear within one or two standard deviations of the baselines. The claim that Term2Note 'consistently outperforms' baselines would be strengthened by significance testing or by reporting effect sizes with confidence intervals; at present, the broad trend is clear, but the magnitude of the utility advantage is not precisely established.
minor comments (5)
  1. [§3.5] The phrase 'we preform preference sampling' contains a typo; it should read 'perform'.
  2. [§3.3, Eq. (2)] Equation (2) and the surrounding text use σ_t for the DPRP* noise scale, but Algorithm 2 parameterizes the noise as σ_1 and σ_2 derived from (ε_1, δ_1) and (ε_2, δ_2). Please reconcile these notations and clarify which privacy split (b=0.85) corresponds to which noise term.
  3. [Table 1 and throughout] The baseline name is written inconsistently as 'FastDP' in the text, 'Fast-DP' in Table 1, and 'Fast-DP' in the supplementary tables. Please use one convention throughout.
  4. [Appendix C, Table 2] The section title 'Discharge Disposition' is missing a closing quote in the table; more generally, the section grouping table would be easier to read if the entries were formatted consistently.
  5. [Appendix F] The membership-inference distance analysis is described as 'preliminary' and provides only an indirect sanity check. The caption of Figure 5 should note that the distance distributions are not a formal privacy guarantee and that the analysis is subject to the limitations the authors themselves mention (e.g., no canary-based lower bound).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: held-out SNOMED test set, real-note test evaluation, and external baselines make the empirical claims self-contained; the DPRP* proof gap is a missing proof, not a circular reduction.

full rationale

The paper's empirical derivation chain is self-contained. TERMGEN is trained on MIMIC-III public terms and NOTEGEN on MIMIC-IV train, while the 204-note SNOMED test subset is explicitly excluded from training; utility is measured by training classifiers on synthetic notes and testing on real notes, so no fitted constant is recycled as a prediction. The DP quality maximiser uses an external clinical perplexity model (Asclepius-Llama3-8B), not the generator itself. Self-citations in the introduction (e.g., Schlegel et al. 2025; Schlegel et al. 2023; Li et al. 2023a) are contextual and not load-bearing for the central claims. The one serious weakness is not circularity: Appendix B asserts 'M2 = TERMGEN processing on D_test, which is (epsilon_t, delta_t)-DP' without proving it from Algorithm 2, and Algorithm 2 provides no sensitivity bound or explicit noise calibration (line 7, 'E_DP = E' V'_k^T + V'_k^T', is also dimensionally ambiguous). That is an unproven privacy guarantee — a correctness/completeness gap — but the paper does not define epsilon_t or delta_t in terms of the final output, nor fit any parameter to the test set. Therefore no circular step meeting the evidentiary standard is present; score 0.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard DP theorems and domain transfer assumptions. The main unproved load-bearing assumption is the DP-ness of the adapted DPRP* mechanism for embeddings. Several hyperparameters are hand-chosen.

free parameters (4)
  • DPRP* privacy split b = 0.85
    Fraction of the per-section privacy budget assigned to embedding noise in Algorithm 2; chosen by hand, no sensitivity analysis given.
  • DPRP* projection dimension ratio = 0.6 * embedding_dim
    Number of SVD components kept in Algorithm 2; chosen by hand.
  • Embedding training noise sigma_emb = 0.05
    Gaussian noise added to embeddings during TERMGEN training to reduce train/inference gap; chosen by hand.
  • DP quality maximiser candidate count k = 4 for Term2Note and FastDP, 7 for AUG-PE
    Number of candidate notes sampled before perplexity selection; chosen by hand.
assumptions (4)
  • standard math Parallel composition and post-processing theorems of DP
    Used in Section 3.6 and Appendix B to combine NOTE-GEN and TERMGEN privacy losses.
  • ad hoc to paper DPRP* as described in Gondara and Wang (2020) remains DP when applied to term embeddings
    The paper asserts this adaptation without proof; no sensitivity or noise calibration for MedEmbed embeddings is provided.
  • domain assumption Public MIMIC-III term lists are representative enough to train a term-generation model that transfers to MIMIC-IV
    TERMGEN is trained only on D_public (MIMIC-III) and applied to private MIMIC-IV terms; transfer is assumed.
  • domain assumption The six-category section taxonomy covers the structure of discharge notes
    SECSPLIT groups section titles into six categories (Appendix C); notes outside this taxonomy would be mis-segmented.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Term2Note: Synthesising Differentially Private Clinical Notes from Medical Terms." pith.science (2026). https://pith.science/paper/CGJGPCD4

@misc{pith2026250910882,
  author       = {Pith},
  title        = {Pith review of: Term2Note: Synthesising Differentially Private Clinical Notes from Medical Terms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CGJGPCD4}},
  note         = {Machine review of arXiv:2509.10882}
}
read the original abstract

Training data is fundamental to the success of modern machine learning models, yet in high-stakes domains such as healthcare, the use of real-world training data is severely constrained by concerns over privacy leakage. A promising solution to this challenge is the use of differentially private (DP) synthetic data, which offers formal privacy guarantees while maintaining data utility. However, striking the right balance between privacy protection and utility remains challenging in clinical note synthesis, given its domain specificity and the complexity of long-form text generation. In this paper, we present Term2Note, a methodology to synthesise long clinical notes under strong DP constraints. By structurally separating content and form, Term2Note generates section-wise note content conditioned on DP medical terms, with each governed by separate DP constraints. A DP quality maximiser further enhances synthetic notes by selecting high-quality outputs. Experimental results show that Term2Note produces synthetic notes with statistical properties closely aligned with real clinical notes, demonstrating strong fidelity. In addition, multi-label classification models trained on these synthetic notes perform comparably to those trained on real data, confirming their high utility. Compared to existing DP text generation baselines, Term2Note achieves substantial improvements in both fidelity and utility while operating under fewer assumptions, suggesting its potential as a viable privacy-preserving alternative to using sensitive clinical notes.

Figures

Figures reproduced from arXiv: 2509.10882 by the authors.

Figure 1
Figure 1. Overview of Term2Note. Given an original clinical note, a section splitter and medical term extractor derive section￾wise medical terms, which are embedded using a clinical text embedder. Depending on whether privatising medical terms, different perturbation mechanisms are applied to enforce DP on the embeddings. Conditioned on the perturbed embeddings, a term generation model produces synthetic section-wise medical… view at source ↗
Figure 2
Figure 2. Distribution of n-gram frequencies in clinical notes [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Human evaluation results summarised using the [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Examples of the last section in clinical notes generated by different models. For illustration purposes, some content is [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Distribution of minimum cosine distances for all evaluated synthetic strategies compared to the baseline of original [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Text length distribution. H.2 Utility [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 27 canonical work pages

  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]

    J.; McMahan, H

    Abadi, M.; Chu, A.; Goodfellow, I. J.; McMahan, H. B.; Mironov, I.; Talwar, K.; and Zhang, L. 2016. Deep Learning with Differential Privacy. In CCS , 308--318. ACM

    work page 2016

  4. [4]

    Adnan, M.; Warren, J.; and Orr, M. 2010. Assessing text characteristics of electronic discharge summaries and their implications for patient readability. In Proceedings of the Fourth Australasian Workshop on Health Informatics and Knowledge Management, volume 108, 77--84

    work page 2010

  5. [5]

    I.; and Mishra, A

    Alzoubi, Y. I.; and Mishra, A. 2025. Differential privacy and artificial intelligence: potentials, challenges, and future avenues. EURASIP J. Inf. Secur. , 2025(1): 18

    work page 2025

  6. [6]

    Balachandran, A. 2024. MedEmbed: Medical-Focused Embedding Models

    work page 2024

  7. [7]

    A.; and Terry, M

    Bradley, R. A.; and Terry, M. E. 1952. Rank analysis of incomplete block designs: I. the method of paired comparisons. Biometrika, 39(3/4): 324--345

    work page 1952

  8. [8]

    Bu, Z.; Wang, Y.; Zha, S.; and Karypis, G. 2023. Differentially Private Optimization on Large Model at Small Cost. In ICML , volume 202 of Proceedings of Machine Learning Research, 3192--3218. PMLR

    work page 2023

Show all 42 references
  1. [9]

    Dwork, C.; McSherry, F.; Nissim, K.; and Smith, A. D. 2006. Calibrating Noise to Sensitivity in Private Data Analysis. In TCC , volume 3876 of Lecture Notes in Computer Science, 265--284. Springer

    2006

  2. [10]

    Dwork, C.; and Roth, A. 2014. The Algorithmic Foundations of Differential Privacy. Found. Trends Theor. Comput. Sci., 9(3-4): 211--407

    2014

  3. [11]

    P.; Dong, H.; Daines, L.; Basetti, S.; Holder, M.; Penfold, R

    Falis, M.; Gema, A. P.; Dong, H.; Daines, L.; Basetti, S.; Holder, M.; Penfold, R. S.; Birch, A.; and Alex, B. 2024. Can GPT-3.5 generate and code discharge summaries? J. Am. Medical Informatics Assoc., 31(10): 2284--2293

    2024

  4. [12]

    Flemings, J.; and Annavaram, M. 2024. Differentially Private Knowledge Distillation via Synthetic Text Generation. In ACL (Findings) , 12957--12968. Association for Computational Linguistics

    2024

  5. [13]

    M.; and Afshar, M

    Gao, Y.; Dligach, D.; Miller, T.; Xu, D.; Churpek, M. M.; and Afshar, M. 2022. Summarizing Patients' Problems from Hospital Progress Notes Using Pre-trained Sequence-to-Sequence Models. In COLING , 2979--2991. International Committee on Computational Linguistics

    2022

  6. [14]

    Gondara, L.; and Wang, K. 2020. Differentially Private Small Dataset Release Using Random Projections. In UAI , volume 124 of Proceedings of Machine Learning Research, 639--648. AUAI Press

    2020

  7. [15]

    Google. 2025. Gemma 3

    2025

  8. [16]

    W.; Ngo, H.; Johnson, A.; and Pollard, T

    Hardman, W.; Banks, M.; Davidson, R.; Truran, D.; Ayuningtyas, N. W.; Ngo, H.; Johnson, A.; and Pollard, T. 2025. SNOMED CT Entity Linking Challenge (version 1.1.0) . https://doi.org/10.13026/qn8t-6e19. PhysioNet. RRID:SCR\_007345

    2025 doi

  9. [17]

    A.; and Mark, R

    Johnson, A.; Pollard, T.; Horng, S.; Celi, L. A.; and Mark, R. 2023. MIMIC-IV-Note: Deidentified free-text clinical notes (version 2.2) . https://doi.org/10.13026/1n74-ne17. PhysioNet. RRID:SCR\_007345

    2023 doi

  10. [18]

    E.; Pollard, T

    Johnson, A. E.; Pollard, T. J.; Shen, L.; Lehman, L.-w. H.; Feng, M.; Ghassemi, M.; Moody, B.; Szolovits, P.; Anthony Celi, L.; and Mark, R. G. 2016. MIMIC-III, a freely accessible critical care database. Scientific data, 3(1): 1--9

    2016

  11. [19]

    B.; Chess, B.; Child, R.; Gray, S.; Radford, A.; Wu, J.; and Amodei, D

    Kaplan, J.; McCandlish, S.; Henighan, T.; Brown, T. B.; Chess, B.; Child, R.; Gray, S.; Radford, A.; Wu, J.; and Amodei, D. 2020. Scaling Laws for Neural Language Models. CoRR, abs/2001.08361

    2020 arXiv

  12. [20]

    Kurakin, A.; Ponomareva, N.; Syed, U.; MacDermed, L.; and Terzis, A. 2023. Harnessing large-language models to generate private synthetic text. CoRR, abs/2306.01684

    2023 arXiv

  13. [21]

    H.; You, S

    Kweon, S.; Kim, J.; Kim, J.; Im, S.; Cho, E.; Bae, S.; Oh, J.; Lee, G.; Moon, J. H.; You, S. C.; Baek, S.; Han, C. H.; Jung, Y. B.; Jo, Y.; and Choi, E. 2024 a . Publicly Shareable Clinical Large Language Model Built on Synthetic Clinical Notes. In ACL (Findings) , 5148--5168....

    2024

  14. [22]

    Kweon, S.; Kim, J.; Kwak, H.; Cha, D.; Yoon, H.; Kim, K.; Yang, J.; Won, S.; and Choi, E. 2024 b . EHRNoteQA: An LLM Benchmark for Real-World Clinical Practice Using Discharge Summaries. In NeurIPS

    2024

  15. [23]

    Labrak, Y.; Bazoge, A.; Morin, E.; Gourraud, P.-A.; Rouvier, M.; and Dufour, R. 2024. BioMistral: A Collection of Open-Source Pretrained Large Language Models for Medical Domains. arXiv:2402.10373

    2024 arXiv

  16. [24]

    Lee, J.; and Kifer, D. 2020. Scaling up Differentially Private Deep Learning with Fast Per-Example Gradient Clipping. CoRR, abs/2009.03106

    2020 arXiv

  17. [25]

    R.; Zeng, X.; Beck, D.; Winkler, S.; and Nenadic, G

    Li, H.; Wu, Y.; Schlegel, V.; Batista - Navarro, R.; Nguyen, T.; Kashyap, A. R.; Zeng, X.; Beck, D.; Winkler, S.; and Nenadic, G. 2023 a . Team: PULSAR at ProbSum 2023: PULSAR: Pre-training with Extracted Healthcare Terms for Summarising Patients' Problems and Data Augmentatio...

    2023

  18. [26]

    M.; Ahmad, F

    Li, Y.; Wehbe, R. M.; Ahmad, F. S.; Wang, H.; and Luo, Y. 2023 b . A comparative study of pretrained language models for long clinical text. Journal of the American Medical Informatics Association, 30(2): 340--347

    2023

  19. [27]

    Mattern, J.; Jin, Z.; Weggenmann, B.; Sch \" o lkopf, B.; and Sachan, M. 2022. Differentially Private Language Models for Secure Data Sharing. In EMNLP , 4860--4873. Association for Computational Linguistics

    2022

  20. [28]

    McSherry, F. 2009. Privacy integrated queries: an extensible platform for privacy-preserving data analysis. In SIGMOD Conference , 19--30. ACM

    2009

  21. [29]

    Peng, B.; Li, C.; He, P.; Galley, M.; and Gao, J. 2023. Instruction Tuning with GPT-4 . CoRR, abs/2304.03277

    2023 arXiv

  22. [30]

    Pillutla, K.; Swayamdipta, S.; Zellers, R.; Thickstun, J.; Welleck, S.; Choi, Y.; and Harchaoui, Z. 2021. MAUVE: Measuring the Gap Between Neural Text and Human Text using Divergence Frontiers. In NeurIPS, 4816--4828

    2021

  23. [31]

    Radford, A.; Wu, J.; Child, R.; Luan, D.; Amodei, D.; Sutskever, I.; et al. 2019. Language models are unsupervised multitask learners. OpenAI blog, 1(8): 9

    2019

  24. [32]

    Schick, T.; and Sch \" u tze, H. 2021. Generating Datasets with Pretrained Language Models. In EMNLP (1) , 6943--6951. Association for Computational Linguistics

    2021

  25. [33]

    A.; Zhao, Z.; and Yee, K

    Schlegel, V.; Bharath, A. A.; Zhao, Z.; and Yee, K. 2025. Generating Synthetic Data with Formal Privacy Guarantees: State of the Art and the Road Ahead. CoRR, abs/2503.20846

    2025 arXiv

  26. [34]

    R.; Beck, D.; Zeng, X.; Batista - Navarro, R

    Schlegel, V.; Li, H.; Wu, Y.; Subramanian, A.; Nguyen, T.; Kashyap, A. R.; Beck, D.; Zeng, X.; Batista - Navarro, R. T.; Winkler, S.; and Nenadic, G. 2023. PULSAR at MEDIQA-Sum 2023: Large Language Models Augmented by Synthetic Dialogue Convert Patient Dialogues to Medical Rec...

    2023

  27. [35]

    Soldaini, L. 2016. QuickUMLS: a fast, unsupervised approach for medical concept extraction

    2016

  28. [36]

    Taori, R.; Gulrajani, I.; Zhang, T.; Dubois, Y.; Li, X.; Guestrin, C.; Liang, P.; and Hashimoto, T. B. 2023. Stanford Alpaca: An Instruction-following LLaMA model. https://github.com/tatsu-lab/stanford_alpaca

    2023

  29. [37]

    U.S. NLM . 2025. UMLS Knowledge Sources [dataset on the Internet]. http://www.nlm.nih.gov/research/umls/licensedcontent/umlsknowledgesources.html. Release 2024AA. Bethesda (MD): National Library of Medicine (US); 2024 May 6 [cited 2024 Jul 15]

    2025

  30. [38]

    successful

    Weetman, K.; Spencer, R.; Dale, J.; Scott, E.; and Schnurr, S. 2021. What makes a “successful” or “unsuccessful” discharge letter? Hospital clinician and General Practitioner assessments of the quality of discharge letters. BMC health services research, 21(1): 349

    2021

  31. [39]

    A.; Nori, H.; Jiang, H.; Zhang, H.; Lee, Y

    Xie, C.; Lin, Z.; Backurs, A.; Gopi, S.; Yu, D.; Inan, H. A.; Nori, H.; Jiang, H.; Zhang, H.; Lee, Y. T.; Li, B.; and Yekhanin, S. 2024. Differentially Private Synthetic Data via Foundation Model APIs 2: Text. In ICML . OpenReview.net

    2024

  32. [40]

    Yang, A.; Li, A.; Yang, B.; Zhang, B.; Hui, B.; Zheng, B.; Yu, B.; Gao, C.; Huang, C.; Lv, C.; Zheng, C.; Liu, D.; Zhou, F.; Huang, F.; Hu, F.; Ge, H.; Wei, H.; Lin, H.; Tang, J.; Yang, J.; Tu, J.; Zhang, J.; Yang, J.; Yang, J.; Zhou, J.; Zhou, J.; Lin, J.; Dang, K.; Bao, K.; ...

    2025 arXiv

  33. [41]

    Yousefpour, A.; Shilov, I.; Sablayrolles, A.; Testuggine, D.; Prasad, K.; Malek, M.; Nguyen, J.; Ghosh, S.; Bharadwaj, A.; Zhao, J.; Cormode, G.; and Mironov, I. 2021. Opacus: User-Friendly Differential Privacy Library in PyTorch. CoRR, abs/2109.12298

    2021 arXiv

  34. [42]

    A.; Li, X.; Kumar, G.; McAnallen, J.; Shajari, H.; Sun, H.; Levitan, D.; and Sim, R

    Yue, X.; Inan, H. A.; Li, X.; Kumar, G.; McAnallen, J.; Shajari, H.; Sun, H.; Levitan, D.; and Sim, R. 2023. Synthetic Text Generation with Differential Privacy: A Simple and Practical Recipe. In ACL (1) , 1321--1342. Association for Computational Linguistics

    2023

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.