Pith. sign in

REVIEW 3 major objections 6 minor 53 references

PatientDx: Merging Large Language Models for Protecting Data-Privacy in Healthcare

T0 review · 3 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read PatientDx shows that merging two off-the-shelf language models—one math-specialized, one instruction-tuned—beats fine-tuned medical LLMs on mortality prediction without ever training on patient records.

desk verdict Test-set lambda tuning makes the headline AUROC gains in-sample, but the privacy-preserving-merging framing is new and salvageable. read the letter →

arxiv 2504.17360 v1 pith:YXN32QGS submitted 2025-04-24 cs.CL

Jose G. Moreno (IRIT-IRIS) , Jesus Lovon (IRIT-IRIS) , M'Rick Robin-Charlet (UT3) , Christine Damase-Michel , Lynda Tamine (IRIT-IRIS) This is my paper · ORCID
classification cs.CL
keywords modelmergingprivacy-preservingLLMMIMIC-IVmortalitySLerphealthcarepredictiondataleakagelanguagenumericalreasoning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that a hospital can build a strong predictive language model without ever fine-tuning on patient data: merge two publicly available base models—one instruction-tuned, one math-specialized—using spherical linear interpolation, and tune only a single scalar weight λ against a performance metric. On the MIMIC-IV in-hospital mortality task, the resulting 8B model reaches 0.6338 AUROC, beating the strongest prior biomedical model (Meditron 7B at 0.5890) and both of its own inputs (around 0.50). The paper also claims the merged model is substantially less prone to patient-data leakage than LoRA-fine-tuned alternatives, and that it transfers to numerical medical questions. The reason a reader would care is that it points to a privacy-preserving alternative to fine-tuning: performance gains without storing or exposing patient records in the training loop.

What carries the argument

The load-bearing mechanism is spherical linear interpolation (SLerp), an angular interpolation between the parameter vectors of two same-architecture LLMs: θ* = [sin((1-λ)Ω)/sin(Ω)]θ1 + [sin(λΩ)/sin(Ω)]θ2, where Ω is the angle between the two weight vectors. PatientDx tunes the single scalar λ (e.g., λ=0.4 for PatientDx 8B) on the task metric without updating any weights. The second essential ingredient is the choice of a math-specialized model (DART-math) as the pivot, which supplies numerical reasoning competence that the instruct/biomedical model lacks, letting the merged model parse numeric EHR features without having seen patient data.

What would settle it

Re-run the PatientDx pipeline with lambda chosen on a dedicated validation split (or nested cross-validation) and evaluate on a test fold never used for selection; if PatientDx 8B's AUROC no longer exceeds Meditron 7B's 0.5890, the headline improvement is a selection artifact.

Watch

Extended reading notes

Core claim

The central discovery is that model merging, not fine-tuning, can be the effective adaptation mechanism for privacy-sensitive clinical prediction. PatientDx 8B combines Llama-3.1-8B-Instruct with DART-math-8B via SLerp at λ=0.4, and the merged parameters alone produce better mortality prediction than either input model, the biomedical specialist Meditron 7B, and fine-tuned baselines, while the DLT metrics show no evidence of memorizing the training set. The authors take this as empirical support for their postulate: with a math-capable pivot model, there exists a merge setting that outperforms the input models on private data and does so without adapting the weights to any patient record. Ablations show the math pivot is the most critical ingredient: removing DART-math costs 21.1% in AUROC for PatientDx 8B, more than removing SLerp (9.0%).

Load-bearing premise

The paper selects the lambda that maximizes performance using the same test partition on which it reports final AUROC, so the claimed gains assume that partition is a genuinely held-out set; if it is not clean, the improvement could be a tuning artifact rather than a property of merging.

Editorial extensions

If this is right

  • Merging can replace fine-tuning for at least some healthcare predictive tasks: PatientDx 8B reaches 0.6338 AUROC on MIMIC-IV mortality without parameter updates on patient data.
  • The math pivot is load-bearing: removing DART-math from the merge drops PatientDx 8B by 21.1% in AUROC, and replacing SLerp with linear interpolation drops it by 9.0%.
  • Merged models are less prone to patient-data leakage than fine-tuned models under the DLT perplexity metrics, with Δ1 near 3.85–3.89 for PatientDx versus negative values for fine-tuned models.
  • PatientDx 8B transfers to numerical medical questions, producing a reasoned dose calculation (166.6 mg/8h) where a biomedical baseline drifts off-task.
  • The framework needs only a performance metric to tune λ, not private labels for training, keeping the interaction between model and patient data one-way.

Reading between the lines

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

  • Editorial inference: the same recipe should transfer to other numeric-heavy clinical prediction tasks (e.g., length-of-stay, readmission, dosing), because the mechanism being exploited—arithmetic competence injected by a math-pivot model—is task-agnostic; the paper only tests mortality.
  • Editorial inference: the privacy claim is comparative, not absolute; PatientDx inherits whatever memorized content its input models already contain, and the DLT metrics measure ease of detecting training data, not immunity to membership-inference attacks, so a stronger privacy guarantee would need a formal leakage definition such as differential privacy.
  • Editorial inference: a cheaper way to test the hypothesis would be to vary the pivotal math model across families (e.g., different math-tuned 7–8B checkpoints) and see whether the AUROC gain tracks the pivot's arithmetic benchmark score; the paper does not report this correlation.
  • Editorial inference: the RRF fusion improvement in the retrieval task was not statistically significant, suggesting the transfer benefit may be limited to tasks requiring numerical computation rather than general retrieval.
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

3 major / 6 minor

Summary. The paper proposes PatientDx, a framework that builds task-specialized LLMs for healthcare prediction by merging publicly available pre-trained models using SLerp or linear interpolation. The proposed configurations combine an instruction-tuned or biomedical model with a math-specialized model (Mathstral or DART-math), and are evaluated on MIMIC-IV mortality and mortality-hard prediction tasks. The paper reports AUROC/AUPRC gains over zero-shot baselines, an ablation of merging components, perplexity-based DLT leakage diagnostics, a qualitative medical question-answering example, and a query-expansion retrieval experiment on ReCDS-PPR. The central claim is that model merging avoids fine-tuning or adaptation on patient data while improving predictive performance and reducing leakage risk.

Significance. If the central claim were supported, the paper would make a useful contribution: model merging would be a lightweight, privacy-preserving alternative to fine-tuning for numerical electronic health record tasks, and the choice of a math-specialized pivot is a concrete, transferable design insight. The authors are also to be credited for releasing the best model, for using a standard merging toolkit, and for including an ablation and a leakage diagnostic that go beyond task accuracy. However, the experimental protocol as reported does not support the generalization claim: the hyperparameters in Table 2 are selected on the same test partition used for the reported scores, so the headline improvement is an in-sample selection artifact. The leakage analysis, while a useful diagnostic, does not measure the influence of the selection procedure on the released model. The core idea remains worth pursuing, but the evidence presented here needs a clean held-out evaluation before the claims can be accepted.

major comments (3)
  1. [§4.1, footnote 2; §5 Limitations] The protocol described in §4.1 and in the Limitations section means that the λ* values in Table 2 are chosen by maximizing AUROC on the exact test partition that is then used to report the headline results. The footnote 'Only in test partition given the computational cost' and the Limitations sentence 'hyper-parameter selection could be performed on the full training set (without k-fold on test as described in §4.1)' make this explicit. Consequently, the PatientDx 8B AUROC of 0.6338 is the value at the best λ on the test fold, not an unbiased estimate of performance on unseen patients, and the comparison against zero-shot, untuned baselines is not a fair evaluation of generalization. This directly undermines the answer to RQ1 and the abstract's claim that the method works without adaptation on patient data. A clean re-evaluation with λ chosen on a training fold and reported on a truly held-out fold, or a nested cross-validation, is required before the improvement claim can be supported.
  2. [§3.2 and Abstract] The statement that PatientDx works 'without requiring fine-tuning nor adaptation on patient data' is not supported by the reported procedure. Section 3.2 says that λ is optimized using a performance metric on examples, and Section 4.1 places that optimization on the test partition. Thus the final model is selected using patient inputs and patient outcomes, even though no gradient updates are performed. The word 'adaptation' in the abstract should be restricted to 'without gradient-based training'; otherwise the privacy motivation is overstated. The DLT metrics in Table 4 assess the released model's perplexity, but they cannot detect the fact that the model's hyperparameters were chosen using the test labels, so they do not establish that the selection process preserved privacy.
  3. [§4.2, Table 2] No uncertainty quantification or significance test is provided for any of the AUROC/AUPRC comparisons. With 6,155 patients and a 10.22% positive rate, the difference between PatientDx 8B (0.6338) and the strongest baseline Meditron 7B (0.5890) is 0.0448 AUROC, and the difference from the input models is about 0.13; neither difference is accompanied by a confidence interval or a test such as DeLong or bootstrap. Several baselines sit at exactly 0.50, which suggests low signal, and the ablation percentages in Table 3 are likewise reported without any indication of variance. Without this, the central 'outperforms' claim is not statistically grounded.
minor comments (6)
  1. [§4.2, Table 2] The sentence 'PatientDx 8B improves by 0.07 absolute points, the strongest baseline' is inconsistent with the table: 0.6338−0.5890=0.0448. The abstract's 'up to 7%' should also be labeled as relative rather than absolute improvement, or corrected.
  2. [Abstract / §4.2] The Hugging Face URL in the abstract points to 'mistral_merged_0_4', but the best model in Table 2 is PatientDx 8B, a Llama-based merge at λ*=0.4; please make sure the released artifact matches the reported configuration and the URL is described correctly.
  3. [Figure 1] Figure 1 is referenced in §3.1 to support Observation 2, but the figure itself is not included in the text; either include it or describe the data behind the curves in prose.
  4. [§4.3, Table 4] The caption says 'rows represent better performance' for the Δ metrics, which is ambiguous because Δ1 and Δ2 move in opposite directions; the caption should state explicitly that larger Δ1 and smaller Δ2 are better.
  5. [§4.4, Table 6] The statement that 'statistical tests show no significance between the two' should name the test used and report the p-value, or be removed; otherwise the claim is not verifiable.
  6. [§4.2, Table 3] The ablation percentages do not state whether the drops are relative or absolute; Table 3 gives e.g. '↓21.1%' for PatientDx 8B w/o Math, but the parenthetical percentages should be defined.

Circularity Check

1 steps flagged · score 6.0 of 10

AUROC gain reduces to test-set hyperparameter selection: λ* is chosen on the same test partition that produces the reported numbers.

  1. fitted input called prediction [Section 3.2 (λ* = argmax m(τ)); Section 4.1 footnote 2; Table 2; Limitations]
    "λ∗ = argmaxλii=1...nm(τ) ... PatientDx requires a metric for optimizing merging hyperparameters such as m(τe)∗≥m(τ)i without training M∗ e on private data ... 2Only in test partition given the computational cost. ... hyper-parameter selection could be performed on the full training set (withoutk-fold on test as described in §4.1) but at significantly higher computational cost."

    The reported AUROC values in Table 2 are obtained after selecting λ* on the same test partition used to compute those AUROC values, because footnote 2 says the 2-fold hyperparameter selection was performed 'Only in test partition'. Therefore each 'PatientDx ... (λ*=...)' row is the maximizer of the evaluation metric evaluated on the evaluation set itself. The claimed improvement over zero-shot baselines is an in-sample selection maximum over the λ grid, not an unbiased estimate on unseen patients, and the baselines did not receive equivalent test-set adaptation. The abstract's 'without requiring fine-tuning nor adaptation on patient data' is contradicted by this protocol: model selection used patient test examples.

full rationale

The paper's central empirical claim is that SLerp-merging Llama3-8B-Instruct with DART-math-8B at λ*=0.4 yields 0.6338 AUROC on MIMIC-IV Mortality, beating Meditron 7B and the input models. The protocol, however, ties the reported number directly to the selection procedure: λ* is defined as argmax of the task metric, and the 2-fold partition used for hyperparameter selection is applied 'Only in test partition'. Consequently, the headline AUROC is not a prediction of performance on held-out patients; it is the value of the selection objective at its chosen optimum. The comparison to zero-shot, untuned baselines is therefore not controlled, since only the merged models received test-set-based adaptation. The Limitations section explicitly acknowledges this by saying generalization would require selection on the full training set 'instead of k-fold on test'. The DLT perplexity analysis and the qualitative case study are independent of this selection issue, but they do not repair the primary AUROC comparison. No other load-bearing circularity was found: the merging equations are standard, and the self-citations to earlier MIMIC-IV work are used only for dataset/prompt configuration, not to justify the central result.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The framework introduces no new particles or forces. The free parameters are the three merging weights, which are fitted to the test set and directly determine the reported performance. The key assumptions are the math-pivot hypothesis and the use of perplexity-based leakage metrics as privacy evidence.

free parameters (3)
  • lambda for PatientDx 7B (Mistral Instruct + Mathstral) = 0.8
    Selected by maximizing AUROC on the test partition (Section 4.1 footnote 2). This value directly controls the reported 0.6057 AUROC result.
  • lambda for PatientDx 8B (Llama Instruct + DART-math) = 0.4
    Selected by maximizing AUROC on the test partition. This is the headline configuration with 0.6338 AUROC.
  • lambda for PatientBioDx 8B (BioMistral + DART-math) = 0.7
    Selected by maximizing AUROC on the test partition. Corresponds to 0.6101 AUROC.
assumptions (3)
  • domain assumption A math-specialized LLM is an effective pivotal model for merging on numerical patient data
    Observation 1 in Section 3.1 postulates that numerical reasoning is key for patient data; the entire design depends on this, with only three merge configurations tested.
  • domain assumption Perplexity differences between train, test, and reference texts (DLT) quantify data leakage or memorization
    Section 4.3 uses DLT as the privacy metric; this is a proxy based on perplexity, not a direct membership-inference or extraction attack.
  • domain assumption Normalized probabilities of only the tokens 'yes' and 'no' provide a valid basis for AUROC and AUPRC computation
    Section 4.1 describes the scoring method, citing Zhuang et al. (2024); the validity of this proxy is assumed throughout the main evaluation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PatientDx: Merging Large Language Models for Protecting Data-Privacy in Healthcare." pith.science (2026). https://pith.science/paper/YXN32QGS

@misc{pith2026250417360,
  author       = {Pith},
  title        = {Pith review of: PatientDx: Merging Large Language Models for Protecting Data-Privacy in Healthcare},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YXN32QGS}},
  note         = {Machine review of arXiv:2504.17360}
}
read the original abstract

Fine-tuning of Large Language Models (LLMs) has become the default practice for improving model performance on a given task. However, performance improvement comes at the cost of training on vast amounts of annotated data which could be sensitive leading to significant data privacy concerns. In particular, the healthcare domain is one of the most sensitive domains exposed to data privacy issues. In this paper, we present PatientDx, a framework of model merging that allows the design of effective LLMs for health-predictive tasks without requiring fine-tuning nor adaptation on patient data. Our proposal is based on recently proposed techniques known as merging of LLMs and aims to optimize a building block merging strategy. PatientDx uses a pivotal model adapted to numerical reasoning and tunes hyperparameters on examples based on a performance metric but without training of the LLM on these data. Experiments using the mortality tasks of the MIMIC-IV dataset show improvements up to 7% in terms of AUROC when compared to initial models. Additionally, we confirm that when compared to fine-tuned models, our proposal is less prone to data leak problems without hurting performance. Finally, we qualitatively show the capabilities of our proposal through a case study. Our best model is publicly available at https://huggingface.co/ Jgmorenof/mistral\_merged\_0\_4.

Figures

Figures reproduced from arXiv: 2504.17360 by the authors.

Figure 1
Figure 1. AUROC performances of Mistral, Llama, and [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 19 canonical work pages

  1. [1]

    Takuya Akiba, Makoto Shing, Yujin Tang, Qi Sun, and David Ha. 2024. Evolutionary optimization of model merging recipes. arXiv preprint arXiv:2403.13187

    arXiv 2024

  2. [2]

    Rohan Anil, Andrew M Dai, Orhan Firat, Melvin Johnson, Dmitry Lepikhin, Alexandre Passos, Siamak Shakeri, Emanuel Taropa, Paige Bailey, Zhifeng Chen, et al. 2023. Palm 2 technical report. arXiv preprint arXiv:2305.10403

    arXiv 2023

  3. [3]

    Elias Bassani. 2022. https://doi.org/10.1007/978-3-030-99739-7\_30 ranx: A blazing-fast python library for ranking evaluation and comparison . In Advances in Information Retrieval - 44th European Conference on IR Research, ECIR 2022, Stavanger, Norway, April 10-14, 2022, Proceedings, Part II , volume 13186 of Lecture Notes in Computer Science, pages 259--...

    doi:10.1007/978-3-030-99739-7 2022

  4. [4]

    Hannah Brown, Katherine Lee, Fatemehsadat Mireshghallah, Reza Shokri, and Florian Tram\` e r. 2022. https://doi.org/10.1145/3531146.3534642 What does it mean for a language model to preserve privacy? In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency, FAccT '22, page 2280–2292, New York, NY, USA. Association for Comput...

    arXiv 2022

  5. [5]

    Zhiqi Bu, Yu-Xiang Wang, Sheng Zha, and George Karypis. 2024. Automatic clipping: differentially private deep learning made easier and stronger. In Proceedings of the 37th International Conference on Neural Information Processing Systems, NIPS '23, Red Hook, NY, USA. Curran Associates Inc

    work page 2024

  6. [6]

    Felix Busch, Lena Hoffmann, Christopher Rueger, Elon HC van Dijk, Rawen Kader, Esteban Ortiz-Prado, Marcus R Makowski, Luca Saba, Martin Hadamitzky, Jakob Nikolas Kather, et al. 2024. Systematic review of large language models for patient care: Current applications and challenges. medRxiv, pages 2024--03

    work page 2024

  7. [7]

    Nicholas Carlini, Daphne Ippolito, Matthew Jagielski, Katherine Lee, Florian Tram \`e r, and Chiyuan Zhang. 2023. https://api.semanticscholar.org/CorpusID:246863735 Quantifying memorization across neural language models . In ICLR'23, volume abs/2202.07646

    arXiv 2023

  8. [8]

    Brown, Dawn Xiaodong Song, \'U lfar Erlingsson, Alina Oprea, and Colin Raffel

    Nicholas Carlini, Florian Tram \`e r, Eric Wallace, Matthew Jagielski, Ariel Herbert-Voss, Katherine Lee, Adam Roberts, Tom B. Brown, Dawn Xiaodong Song, \'U lfar Erlingsson, Alina Oprea, and Colin Raffel. 2020. https://api.semanticscholar.org/CorpusID:229156229 Extracting training data from large language models . In USENIX Security Symposium

    2020

Show all 53 references
  1. [9]

    Zeming Chen, Alejandro Hernández Cano, Angelika Romanou, Antoine Bonnet, Kyle Matoba, Francesco Salvi, Matteo Pagliardini, Simin Fan, Andreas Köpf, Amirkeivan Mohtashami, Alexandre Sallinen, Alireza Sakhaeirad, Vinitra Swamy, Igor Krawczuk, Deniz Bayazit, Axel Marmet, Syrielle...

    2023 arXiv

  2. [10]

    Clément Christophe, Praveen K Kanithi, Tathagata Raha, Shadab Khan, and Marco AF Pimentel. 2024. https://arxiv.org/abs/2408.06142 Med42-v2: A suite of clinical llms . Preprint, arXiv:2408.06142

    2024 arXiv

  3. [11]

    MohammadReza Davari and Eugene Belilovsky. 2023. Model breadcrumbs: Scaling multi-task model merging with sparse masks. arXiv preprint arXiv:2312.06795

    2023 arXiv

  4. [12]

    Jingcheng Du, Yang Xiang, Madhuri Sankaranarayanapillai, Meng Zhang, Jingqi Wang, Yuqi Si, Huy Anh Pham, Hua Xu, Yong Chen, and Cui Tao. 2021. Extracting postmarketing adverse events from safety reports in the vaccine adverse event reporting system (vaers) using deep learning....

    2021

  5. [13]

    John W Ely, Jerome A Osheroff, Mark H Ebell, George R Bergus, Barcey T Levy, M Lee Chambliss, and Eric R Evans. 1999. Analysis of questions asked by family doctors regarding patient care. Bmj, 319(7206):358--361

    1999

  6. [14]

    Charles Goddard, Shamane Siriwardhana, Malikeh Ehghaghi, Luke Meyers, Vlad Karpukhin, Brian Benedict, Mark McQuade, and Jacob Solawetz. 2024. Arcee's mergekit: A toolkit for merging large language models. arXiv preprint arXiv:2403.13257

    2024 arXiv

  7. [15]

    Stefan Hegselmann, Alejandro Buendia, Hunter Lang, Monica Agrawal, Xiaoyi Jiang, and David A. Sontag. 2022. https://api.semanticscholar.org/CorpusID:252992811 Tabllm: Few-shot classification of tabular data with large language models . In AISTATG, volume abs/2210.10723

    2022 arXiv

  8. [16]

    Wang, Chenhui Zhang, Zhangheng Li, Bo Li, and Zhangyang Wang

    Junyuan Hong, Jiachen T. Wang, Chenhui Zhang, Zhangheng Li, Bo Li, and Zhangyang Wang. 2024. Dp-opt: Make large language model your privacy-preserving prompt engineer

    2024

  9. [17]

    Yu, and Xuyun Zhang

    Hongsheng Hu, Zoran Salcic, Lichao Sun, Gillian Dobbie, Philip S. Yu, and Xuyun Zhang. 2022. https://doi.org/10.1145/3523273 Membership inference attacks on machine learning: A survey . ACM Comput. Surv., 54(11s)

    2022 doi

  10. [18]

    Gabriel Ilharco, Marco Tulio Ribeiro, Mitchell Wortsman, Ludwig Schmidt, Hannaneh Hajishirzi, and Ali Farhadi. 2022. Editing models with task arithmetic. In The Eleventh International Conference on Learning Representations

    2022

  11. [19]

    Young Kyun Jang, Dat Huynh, Ashish Shah, Wen-Kai Chen, and Ser-Nam Lim. 2024. https://api.semanticscholar.org/CorpusID:269484503 Spherical linear interpolation and text-anchoring for zero-shot composed image retrieval . ArXiv, abs/2405.00571

    2024 arXiv

  12. [20]

    Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas ...

    2023 arXiv

  13. [21]

    Alistair E. W. Johnson, Lucas Bulgarelli, Lu Shen, Alvin Gayles, Ayad Shammout, Steven Horng, Tom J. Pollard, Benjamin Moody, Brian Gow, Li-wei H. Lehman, Leo A. Celi, and Roger G. Mark. 2023. https://doi.org/10.1038/s41597-022-01899-x Mimic-iv, a freely accessible electronic ...

    2023 doi

  14. [22]

    Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, Arjun Nitin Bhagoji, Kallista Bonawit, Zachary Charles, Graham Cormode, Rachel Cummings, Rafael G

    Peter Kairouz, H. Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, Arjun Nitin Bhagoji, Kallista Bonawit, Zachary Charles, Graham Cormode, Rachel Cummings, Rafael G. L. D’Oliveira, Hubert Eichner, Salim El Rouayheb, David Evans, Josh Gardner, Zachary Garrett, Adr...

    2021

  15. [23]

    Nikhil Kandpal, Eric Wallace, and Colin Raffel. 2022. https://api.semanticscholar.org/CorpusID:246823128 Deduplicating training data mitigates privacy risks in language models . ArXiv

    2022

  16. [24]

    Yanis Labrak, Adrien Bazoge, Emmanuel Morin, Pierre-Antoine Gourraud, Mickael Rouvier, and Richard Dufour. 2024. https://aclanthology.org/2024.findings-acl.348 B io M istral: A collection of open-source pretrained large language models for medical domains . In Findings of the ...

    2024

  17. [25]

    Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chan Ho So, and Jaewoo Kang. 2020. Biobert: a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics, 36(4):1234--1240

    2020

  18. [26]

    Hashimoto

    Xuechen Li, Florian Tram \`e r, Percy Liang, and Tatsunori B. Hashimoto. 2021. https://api.semanticscholar.org/CorpusID:238634219 Large language models can be strong differentially private learners . ArXiv

    2021

  19. [27]

    Moreno, Christine Damase-Michel, and Lynda Tamine

    Jesus Lovon, Martin Mouysset, Jo Oleiwan, Jose G. Moreno, Christine Damase-Michel, and Lynda Tamine. 2025. https://arxiv.org/abs/2501.09384 Evaluating llm abilities to understand tabular electronic health records: A comprehensive study of patient data extraction and retrieval ...

    2025 arXiv

  20. [28]

    Moreno, and Lynda Tamine

    Jesus Lovon-Melgarejo, Thouria Ben-Haddi, Jules Di Scala, Jose G. Moreno, and Lynda Tamine. 2024. https://aclanthology.org/2024.cl4health-1.23 Revisiting the MIMIC - IV benchmark: Experiments using language models for electronic health records . In Proceedings of the First Wor...

    2024

  21. [29]

    Alexandra Sasha Luccioni, Sylvain Viguier, and Anne-Laure Ligozat. 2023. Estimating the carbon footprint of bloom, a 176b parameter language model. Journal of Machine Learning Research, 24(253):1--15

    2023

  22. [30]

    Michael S Matena and Colin A Raffel. 2022. Merging models with fisher-weighted averaging. Advances in Neural Information Processing Systems, 35:17703--17716

    2022

  23. [31]

    H. B. McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Ag \"u era y Arcas. 2016. https://api.semanticscholar.org/CorpusID:14955348 Communication-efficient learning of deep networks from decentralized data . In International Conference on Artificial Intelligence an...

    2016

  24. [32]

    Seth Neel and Peter Chang. 2023. Privacy issues in large language models: A survey. arXiv preprint arXiv:2312.06717

    2023 arXiv

  25. [33]

    Guillermo Ortiz-Jimenez, Alessandro Favero, and Pascal Frossard. 2024. Task arithmetic in the tangent space: Improved editing of pre-trained models. Advances in Neural Information Processing Systems, 36

    2024

  26. [34]

    Ahmed Salem, Apratim Bhattacharya, Michael Backes, Mario Fritz, and Yang Zhang. 2020. Updates-leak: data set inference and reconstruction attacks in online learning. In Proceedings of the 29th USENIX Conference on Security Symposium, SEC'20, USA. USENIX Association

    2020

  27. [35]

    Inan, Amir Houmansadr, and Robert Sim

    Virat Shejwalkar, Huseyin A. Inan, Amir Houmansadr, and Robert Sim. 2021. Membership inference attacks against nlp classification models. In Proceedings NeurIPS 2021 Workshop PRIML

    2021

  28. [36]

    Karan Singhal, Tao Tu, Juraj Gottweis, Rory Sayres, Ellery Wulczyn, Le Hou, Kevin Clark, Stephen Pfohl, Heather Cole-Lewis, Darlene Neal, et al. 2023. Towards expert-level medical question answering with large language models. arXiv preprint arXiv:2305.09617

    2023 arXiv

  29. [37]

    Robin Staab, Mark Vero, Mislav Balunovi'c, and Martin T. Vechev. 2024. https://api.semanticscholar.org/CorpusID:263834989 Beyond memorization: Violating privacy via inference with large language models . In ICLR'24

    2024

  30. [38]

    Xinyu Tang, Richard Shin, Huseyin Inan, Andre Manoel, Fatemehsadat Mireshghallah, Zinan Lin, Sivakanth Gopi, Janardhan (Jana) Kulkarni, and Robert Sim. 2024. Privacy-preserving in-context learning with differentially private few-shot generation. In ICLR 2024

    2024

  31. [39]

    Yuxuan Tong, Xiwen Zhang, Rui Wang, Ruidong Wu, and Junxian He. 2024. https://arxiv.org/abs/2407.13690 Dart-math: Difficulty-aware rejection tuning for mathematical problem-solving

    2024 arXiv

  32. [40]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. https://arxiv.org/abs/2302.13971 Llama:...

    2023 arXiv

  33. [41]

    Tianwen Wei, Liang Zhao, Lichang Zhang, Bo Zhu, Lijie Wang, Haihua Yang, Biye Li, Cheng Cheng, Weiwei Lü, Rui Hu, Chenxia Li, Liu Yang, Xilin Luo, Xuejie Wu, Lunan Liu, Wenjun Cheng, Peng Cheng, Jianhao Zhang, Xiaoyu Zhang, Lei Lin, Xiaokun Wang, Yutuan Ma, Chuanhai Dong, Yanq...

    2023 arXiv

  34. [42]

    BigScience Workshop, Teven Le Scao, Angela Fan, Christopher Akiki, Ellie Pavlick, Suzana Ili \'c , Daniel Hesslow, Roman Castagn \'e , Alexandra Sasha Luccioni, Fran c ois Yvon, et al. 2022. Bloom: A 176b-parameter open-access multilingual language model. arXiv preprint arXiv:...

    2022 arXiv

  35. [43]

    Mitchell Wortsman, Gabriel Ilharco, Samir Ya Gadre, Rebecca Roelofs, Raphael Gontijo-Lopes, Ari S Morcos, Hongseok Namkoong, Ali Farhadi, Yair Carmon, Simon Kornblith, et al. 2022. Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasin...

    2022

  36. [44]

    Lingling Xu, Haoran Xie, Si-Zhao Joe Qin, Xiaohui Tao, and Fu Lee Wang. 2023. Parameter-efficient fine-tuning methods for pretrained language models: A critical review and assessment. arXiv preprint arXiv:2312.12148

    2023 arXiv

  37. [45]

    Biwei Yan, Kun Li, Minghui Xu, Yueyan Dong, Yue Zhang, Zhaochun Ren, and Xiuzheng Cheng. 2024. On protecting the data privacy of large language models (llms): A survey. arXiv preprint arXiv:2403.05156

    2024 arXiv

  38. [46]

    Rui Ye, Wenhao Wang, Jingyi Chai, Dihan Li, Zexi Li, Yinda Xu, Yaxin Du, Yanfeng Wang, and Siheng Chen. 2024. https://doi.org/10.1145/3637528.3671582 Openfedllm: Training large language models on decentralized private data via federated learning . In Proceedings of the 30th AC...

    2024

  39. [47]

    Xiang Yue, Huseyin Inan, Xuechen Li, Girish Kumar, Julia McAnallen, Hoda Shajari, Huan Sun, David Levitan, and Robert Sim. 2023. https://doi.org/10.18653/v1/2023.acl-long.74 Synthetic text generation with differential privacy: A simple and practical recipe . In Proceedings of ...

    2023 doi

  40. [48]

    Xuandong Zhao, Lei Li, and Yu-Xiang Wang. 2022. https://doi.org/10.18653/v1/2022.naacl-main.69 Provably confidential language modelling . In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Techno...

    2022 doi

  41. [49]

    Zhengyun Zhao, Qiao Jin, Fangyuan Chen, Tuorui Peng, and Sheng Yu. 2023. A large-scale dataset of patient summaries for retrieval-based clinical decision support systems. Scientific data, 10(1):909

    2023

  42. [50]

    Honglei Zhuang, Zhen Qin, Kai Hui, Junru Wu, Le Yan, Xuanhui Wang, and Michael Bendersky. 2024. https://doi.org/10.18653/v1/2024.naacl-short.31 Beyond yes and no: Improving zero-shot LLM rankers via scoring fine-grained relevance labels . In Proceedings of the 2024 Conference ...

    2024 doi

  43. [51]

    Max Zimmer, Christoph Spiegel, and Sebastian Pokutta. 2024. Sparse model soups: A recipe for improved pruning via model averaging

    2024

  44. [52]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  45. [53]

    write newline

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

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.