The reviewed record of science sign in
Pith

arxiv: 2606.08769 · v1 · pith:YFOEBSO4 · submitted 2026-06-07 · cs.CL · cs.AI

RadOT-Eval: Auditable Structured-Evidence Transport for Radiology Report Evaluation

Weixin Liu , Juming Xiong , Yang Li , Qingyuan Song , Susannah Rose , Murat Kantarcioglu , Bradley Malin , Zhijun Yin This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-27 18:33 UTCgrok-4.3pith:YFOEBSO4record.jsonopen to challenge →

classification cs.CL cs.AI
keywords radiology report evaluationoptimal transportstructured evidenceerror burdenclinical text generationauditable evaluationSpearman correlation
0
0 comments X

The pith

RadOT-Eval aligns attribute-structured clinical evidence units via entropy-regularized optimal transport to predict error burden more accurately than standard metrics or LLM evaluators.

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

The paper shows that radiology reports can be decomposed into attribute-structured clinical evidence units, then aligned across reference and candidate versions using entropy-regularized optimal transport. Discrepancies in those alignments feed a monotone risk model that estimates total, clinically significant, and insignificant error burden. On an independent test set the method reaches Spearman correlations of 0.715, 0.548, and 0.399 respectively, beating both conventional metrics and an open-source LLM evaluator. The same frozen system also distinguishes corrupted from clean reports at 0.768 AUROC. Readers care because high-stakes clinical text generation needs auditable, rank-oriented scoring that tracks human judgments of factual error rather than surface similarity alone.

Core claim

RadOT-Eval decomposes reference and candidate reports into attribute-structured clinical evidence units, aligns corresponding evidence using entropy-regularized optimal transport, and uses clinically meaningful side-channel discrepancies in a monotone risk model to predict error burden. All transport, feature, and readout choices are selected using the ReXVal dataset, and the frozen system is evaluated on the independent RadEvalX dataset, where it records the reported correlations and stress-test performance.

What carries the argument

Entropy-regularized optimal transport applied to attribute-structured clinical evidence units extracted from radiology reports, which produces alignments whose discrepancy measures enter a risk model for error-burden prediction.

If this is right

  • The approach records higher point estimates of correlation with annotated error burden than standard surface-overlap metrics.
  • It records higher point estimates than the open-source LLM-based evaluator GREEN-radllama2-7B on the same three error-burden categories.
  • In a frozen corruption-sensitivity test it records 0.768 AUROC and a 0.990 win rate for corrupted reports over clean ones.
  • All performance numbers hold under model selection performed only on ReXVal followed by frozen evaluation on RadEvalX.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The explicit evidence alignments could let developers trace which specific omitted findings or polarity reversals drive a high error score.
  • The same unit-extraction and transport steps might transfer to evaluation of other structured clinical narratives such as discharge summaries.
  • If the monotone risk model remains stable, the method could support automated flagging of generated reports whose predicted error burden exceeds a safety threshold.
  • The rank-oriented nature of the transport cost could complement existing pairwise preference datasets for training report generators.

Load-bearing premise

The chosen decomposition of reports into clinical evidence units together with the transport and readout parameters tuned on one dataset will capture the clinically relevant error types on new, independent reports without further adjustment.

What would settle it

On a fresh collection of radiology reports with human error-burden annotations, RadOT-Eval would produce lower Spearman correlation with total error burden than BLEU, ROUGE, or the compared LLM evaluator.

Figures

Figures reproduced from arXiv: 2606.08769 by Bradley Malin, Juming Xiong, Murat Kantarcioglu, Qingyuan Song, Susannah Rose, Weixin Liu, Yang Li, Zhijun Yin.

Figure 1
Figure 1. Figure 1: Overview of RadOT-Eval. Reports are parsed into structured clinical units. Stable attributes define the OT ground cost and transport plan, while [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
read the original abstract

Automatic evaluation is critical for high-stakes text generation, where errors often involve omitted findings, hallucinated content, polarity reversals, location changes, uncertainty mismatches, and temporal-comparison errors rather than low surface similarity alone. Radiology report generation provides a challenging test case because generated reports must preserve structured clinical evidence across sources. We present RadOT-Eval, an interpretable structured-evidence optimal transport framework for offline auditing of radiology report generation. RadOT-Eval decomposes reference and candidate reports into attribute-structured clinical evidence units, aligns corresponding evidence using entropy-regularized optimal transport, and uses clinically meaningful side-channel discrepancies in a monotone risk model to predict error burden. All transport, feature, and readout choices are selected using the ReXVal dataset, and the frozen system is evaluated on the independent RadEvalX dataset. RadOT-Eval achieves Spearman correlations of 0.715, 0.548, and 0.399 with total, clinically significant, and clinically insignificant annotated error burden, respectively, yielding higher point estimates than standard evaluation metrics and the open-source large language model (LLM)-based evaluator GREEN-radllama2-7B. In a frozen auxiliary corruption-sensitivity stress test on ReXErr-v1, RadOT-Eval achieves 0.768 AUROC and a 0.990 corrupted-greater-than-clean paired win rate. These results show that structured evidence transport provides an auditable, rank-oriented evaluation tool for high-stakes generated clinical text under ReXVal-only model selection and frozen RadEvalX testing.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper presents RadOT-Eval, an interpretable framework for evaluating radiology report generation that decomposes reports into attribute-structured clinical evidence units, aligns them via entropy-regularized optimal transport, and applies a monotone risk model using side-channel discrepancies to predict error burden. All transport, feature, and readout choices are selected on the ReXVal dataset with the system frozen for evaluation on the independent RadEvalX dataset. It reports Spearman correlations of 0.715/0.548/0.399 with total/clinically-significant/clinically-insignificant annotated error burden (outperforming standard metrics and GREEN-radllama2-7B) and 0.768 AUROC / 0.990 win-rate on a ReXErr-v1 corruption-sensitivity test.

Significance. If the central results hold under the reported protocol, the work supplies an auditable, rank-oriented alternative to LLM-based evaluators that directly targets structured clinical evidence mismatches rather than surface similarity. The explicit separation of tuning (ReXVal) from frozen testing (RadEvalX) plus the auxiliary corruption stress test are positive design elements that strengthen the case for clinical relevance.

major comments (2)
  1. [Abstract] Abstract (model-selection paragraph): the headline Spearman correlations and AUROC rest on the claim that attribute definitions, entropy-regularization strength, feature choices, and monotone readout—selected on ReXVal—transfer without adjustment to RadEvalX. No quantitative sensitivity analysis (e.g., ablation of the tuning step or alternative attribute schemas) is described, leaving open the possibility that reported gains are tied to ReXVal-specific annotation style rather than general clinical error types.
  2. [Abstract] Abstract (evaluation protocol): while the abstract states that evaluation is frozen after ReXVal selection, the absence of any reported check on whether the discrepancy weights or transport parameters were fitted using information that leaks into the RadEvalX numbers undermines the independence claim and makes it impossible to assess whether the superiority over baselines is robust.
minor comments (1)
  1. [Abstract] The abstract would benefit from a brief parenthetical note on the number of reports in each split and whether error-burden annotations were performed by multiple raters.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on the model-selection and evaluation-protocol sections. We respond point by point below and propose targeted revisions to address the concerns while preserving the manuscript's core claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract (model-selection paragraph): the headline Spearman correlations and AUROC rest on the claim that attribute definitions, entropy-regularization strength, feature choices, and monotone readout—selected on ReXVal—transfer without adjustment to RadEvalX. No quantitative sensitivity analysis (e.g., ablation of the tuning step or alternative attribute schemas) is described, leaving open the possibility that reported gains are tied to ReXVal-specific annotation style rather than general clinical error types.

    Authors: We acknowledge that the current version does not report a quantitative sensitivity analysis of the ReXVal-selected choices. In revision we will add a dedicated ablation subsection that varies attribute schema definitions and regularization strength while measuring the resulting Spearman correlations on the frozen RadEvalX test set. This addition will directly test whether performance depends on ReXVal-specific annotation conventions. revision: yes

  2. Referee: [Abstract] Abstract (evaluation protocol): while the abstract states that evaluation is frozen after ReXVal selection, the absence of any reported check on whether the discrepancy weights or transport parameters were fitted using information that leaks into the RadEvalX numbers undermines the independence claim and makes it impossible to assess whether the superiority over baselines is robust.

    Authors: The manuscript protocol states that all transport, feature, and readout selections occur on ReXVal only, after which the system is frozen. To make the absence of leakage explicit, we will expand the methods section with a numbered enumeration of the exact selection steps, each annotated to confirm that no RadEvalX samples or labels were accessed. This clarification will allow readers to verify the independence claim without altering the reported numbers. revision: yes

Circularity Check

0 steps flagged

ReXVal model selection with frozen independent RadEvalX evaluation exhibits no circularity

full rationale

The paper states that transport/feature/readout choices are selected on ReXVal and the system is frozen for evaluation on independent RadEvalX, with reported Spearman correlations (0.715/0.548/0.399) and AUROC (0.768) computed on the held-out set. No equations, self-citations, or derivations reduce the test-set metrics to the selection data by construction. The central claim rests on an external test distribution rather than any fitted-input-called-prediction or self-definitional pattern.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

Abstract-only review supplies no equations or sections to audit; the framework implicitly rests on the assumption that evidence-unit decomposition is lossless and that ReXVal-tuned choices transfer to RadEvalX.

free parameters (2)
  • entropy regularization strength
    Chosen during ReXVal model selection; value not stated
  • feature and readout choices
    All transport, feature, and readout choices selected using ReXVal
axioms (2)
  • domain assumption Decomposition of reports into attribute-structured clinical evidence units preserves all clinically relevant information
    Central to the alignment step described in the abstract
  • domain assumption Side-channel discrepancies feed a monotone risk model that correctly ranks error burden
    Used to produce the final error-burden prediction

pith-pipeline@v0.9.1-grok · 5834 in / 1653 out tokens · 26053 ms · 2026-06-27T18:33:09.947889+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 2 canonical work pages · 1 internal anchor

  1. [1]

    On faithfulness and factuality in abstractive summarization,

    J. Maynez, S. Narayan, B. Bohnet, and R. McDonald, “On faithfulness and factuality in abstractive summarization,” inProceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020, pp. 1906–1919

    2020

  2. [2]

    Evaluating the factual consistency of abstractive text summarization,

    W. Kry ´sci´nski, B. McCann, C. Xiong, and R. Socher, “Evaluating the factual consistency of abstractive text summarization,” inProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, 2020, pp. 9332–9346. TABLE IX PARSER SCHEMA ANDRADEVALXERROR-CATEGORY DEFINITIONS. Clinical-unit schema span_textVerbatim report span support...

    2020

  3. [3]

    Understanding factuality in abstractive summarization with FRANK: A benchmark for factuality metrics,

    A. Pagnoni, V . Balachandran, and Y . Tsvetkov, “Understanding factuality in abstractive summarization with FRANK: A benchmark for factuality metrics,” inProceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2021, pp. 4812–4829

    2021

  4. [4]

    On the automatic generation of medical imaging reports,

    B. Jing, P. Xie, and E. P. Xing, “On the automatic generation of medical imaging reports,” inProceedings of the 56th Annual Meeting of the Association for Computational Linguistics, 2018, pp. 2577–2586

    2018

  5. [5]

    Clinically accurate chest x-ray report generation,

    G. Liu, T.-M. H. Hsu, M. McDermott, W. Boag, W.-H. Weng, P. Szolovits, and M. Ghassemi, “Clinically accurate chest x-ray report generation,” inProceedings of Machine Learning Research, vol. 106, 2019, pp. 249–269

    2019

  6. [6]

    Improving factual completeness and consistency of image-to-text radiology report generation,

    Y . Miura, Y . Zhang, E. Tsai, C. Langlotz, and D. Jurafsky, “Improving factual completeness and consistency of image-to-text radiology report generation,” inProceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2021, pp. 5288–5304

    2021

  7. [7]

    Evaluating progress in automatic chest x-ray radiology report generation,

    F. Yu, M. Endo, R. Krishnan, I. Pan, A. Tsai, E. P. Reis, E. K. U. N. Fonseca, H. M. H. Lee, Z. S. H. Abad, A. Y . Ng, C. P. Langlotz, V . K. Venugopal, and P. Rajpurkar, “Evaluating progress in automatic chest x-ray radiology report generation,”Patterns, vol. 4, no. 9, p. 100802, 2023

    2023

  8. [8]

    BLEU: a method for automatic evaluation of machine translation,

    K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu, “BLEU: a method for automatic evaluation of machine translation,” inProceedings of the 40th Annual Meeting of the Association for Computational Linguistics, 2002, pp. 311–318

    2002

  9. [9]

    ROUGE: A package for automatic evaluation of summaries,

    C.-Y . Lin, “ROUGE: A package for automatic evaluation of summaries,” inText Summarization Branches Out, 2004, pp. 74–81

    2004

  10. [10]

    BERTScore: Evaluating text generation with BERT,

    T. Zhang, V . Kishore, F. Wu, K. Q. Weinberger, and Y . Artzi, “BERTScore: Evaluating text generation with BERT,” inInternational Conference on Learning Representations, 2020

    2020

  11. [11]

    Radiology Report Generation Models Evaluation Dataset For Chest X-rays (RadEvalX),

    A. R. Calamida, F. Nooralahzadeh, M. Rohanian, M. Nishio, K. Fu- jimoto, and M. Krauthammer, “Radiology Report Generation Models Evaluation Dataset For Chest X-rays (RadEvalX),”PhysioNet, 2024, version 1.0.0

    2024

  12. [12]

    BLEURT: Learning robust metrics for text generation,

    T. Sellam, D. Das, and A. P. Parikh, “BLEURT: Learning robust metrics for text generation,” inProceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020, pp. 7881–7892

    2020

  13. [13]

    MoverScore: Text generation evaluating with contextualized embed- dings and earth mover distance,

    W. Zhao, M. Peyrard, F. Liu, Y . Gao, C. M. Meyer, and S. Eger, “MoverScore: Text generation evaluating with contextualized embed- dings and earth mover distance,” inProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, 2019, pp. 563–578

    2019

  14. [14]

    COMET: A neural framework for MT evaluation,

    R. Rei, C. Stewart, A. C. Farinha, and A. Lavie, “COMET: A neural framework for MT evaluation,” inProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, 2020, pp. 2685– 2702

    2020

  15. [15]

    BARTScore: Evaluating generated text as text generation,

    W. Yuan, G. Neubig, and P. Liu, “BARTScore: Evaluating generated text as text generation,” inAdvances in Neural Information Processing Systems, vol. 34, 2021, pp. 27 263–27 277

    2021

  16. [16]

    Combining automatic labelers and expert annotations for accurate radiology report labeling using BERT,

    A. Smit, S. Jain, P. Rajpurkar, A. Pareek, A. Y . Ng, and M. P. Lungren, “Combining automatic labelers and expert annotations for accurate radiology report labeling using BERT,” inProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, 2020, pp. 1500–1519

    2020

  17. [17]

    RadGraph: Extracting clinical entities and relations from radiology reports,

    S. Jain, A. Agrawal, A. Saporta, S. Q. H. Truong, D. N. Duong, T. Bui, P. Chambon, Y . Zhang, M. P. Lungren, A. Y . Ng, C. P. Langlotz, and P. Rajpurkar, “RadGraph: Extracting clinical entities and relations from radiology reports,” inProceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks, vol. 1, 2021

    2021

  18. [18]

    RadGraph-XL: A large-scale expert-annotated dataset for entity and relation extraction from radiology reports,

    J.-B. Delbrouck, P. Chambon, Z. Chen, M. Varma, A. Johnston, L. Blankemeier, D. Van Veen, T. Bui, S. Truong, and C. Langlotz, “RadGraph-XL: A large-scale expert-annotated dataset for entity and relation extraction from radiology reports,” inFindings of the Associa- tion for Computational Linguistics: ACL 2024, 2024, pp. 12 902–12 915

    2024

  19. [19]

    GREEN: Generative radiology report evaluation and error notation,

    S. Ostmeier, J. Xu, Z. Chen, M. Varma, L. Blankemeier, C. Bluethgen, A. E. Michalson, M. Moseley, C. Langlotz, A. S. Chaudhari, and J.-B. Delbrouck, “GREEN: Generative radiology report evaluation and error notation,” inFindings of the Association for Computational Linguistics: EMNLP 2024, 2024, pp. 374–390

    2024

  20. [20]

    RaTEScore: A metric for radiology report generation,

    W. Zhao, C. Wu, X. Zhang, Y . Zhang, Y . Wang, and W. Xie, “RaTEScore: A metric for radiology report generation,” inProceedings TABLE XII REPRESENTATIVE HIGH-RISK TRANSPORT EDGES FROM QUALITATIVE RECONSTRUCTIONS. MWDENOTES MASS-WEIGHTED;EDGES ARE SELECTED FOR EXPLANATION ONLY AND ARE NOT USED FOR MODEL SELECTION. Case Rank Reference unit Candidate unit Ma...

    2024

  21. [21]

    Radiology Report Expert Evaluation (ReXVal) Dataset,

    F. Yu, M. Endo, R. Krishnan, I. Pan, A. Tsai, E. P. Reis, E. Kaiser Ururahy Nunes Fonseca, H. Lee, Z. Shakeri, A. Ng, C. Langlotz, V . K. Venugopal, and P. Rajpurkar, “Radiology Report Expert Evaluation (ReXVal) Dataset,”PhysioNet, 2023, version 1.0.0

    2023

  22. [22]

    ReX- Err: Synthesizing clinically meaningful errors in diagnostic radiology reports,

    V . M. Rao, S. Zhang, J. N. Acosta, S. Adithan, and P. Rajpurkar, “ReX- Err: Synthesizing clinically meaningful errors in diagnostic radiology reports,”arXiv preprint arXiv:2409.10829, 2024

    work page arXiv 2024

  23. [23]

    ReXErr-v1: Clinically meaningful chest x-ray report errors derived from MIMIC- CXR,

    V . Rao, S. Zhang, J. Acosta, S. Adithan, and P. Rajpurkar, “ReXErr-v1: Clinically meaningful chest x-ray report errors derived from MIMIC- CXR,”PhysioNet, 2025, version 1.0.0

    2025

  24. [24]

    METEOR: An automatic metric for MT evaluation with improved correlation with human judgments,

    S. Banerjee and A. Lavie, “METEOR: An automatic metric for MT evaluation with improved correlation with human judgments,” inPro- ceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, 2005, pp. 65– 72

    2005

  25. [25]

    CIDEr: Consensus-based image description evaluation,

    R. Vedantam, C. L. Zitnick, and D. Parikh, “CIDEr: Consensus-based image description evaluation,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015, pp. 4566–4575

    2015

  26. [26]

    SPICE: Semantic propositional image caption evaluation,

    P. Anderson, B. Fernando, M. Johnson, and S. Gould, “SPICE: Semantic propositional image caption evaluation,” inComputer Vision – ECCV 2016, 2016, pp. 382–398

    2016

  27. [27]

    MAUVE: Measuring the gap between neural text and human text using divergence frontiers,

    K. Pillutla, S. Swayamdipta, R. Zellers, J. Thickstun, S. Welleck, Y . Choi, and Z. Harchaoui, “MAUVE: Measuring the gap between neural text and human text using divergence frontiers,” inAdvances in Neural Information Processing Systems, vol. 34, 2021, pp. 4816–4828

    2021

  28. [28]

    QuestEval: Summarization asks for fact-based evalu- ation,

    T. Scialom, P.-A. Dray, S. Lamprier, B. Piwowarski, J. Staiano, A. Wang, and P. Gallinari, “QuestEval: Summarization asks for fact-based evalu- ation,” inProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 2021, pp. 6594–6604

    2021

  29. [29]

    TieNet: Text- image embedding network for common thorax disease classification and reporting in chest x-rays,

    X. Wang, Y . Peng, L. Lu, Z. Lu, and R. M. Summers, “TieNet: Text- image embedding network for common thorax disease classification and reporting in chest x-rays,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 9049–9058

    2018

  30. [30]

    Generating radiology reports via memory-driven transformer,

    Z. Chen, Y . Song, T.-H. Chang, and X. Wan, “Generating radiology reports via memory-driven transformer,” inProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, 2020, pp. 1439–1449

    2020

  31. [31]

    Cross-modal memory networks for radiology report generation,

    Z. Chen, Y . Shen, Y . Song, and X. Wan, “Cross-modal memory networks for radiology report generation,” inProceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, 2021, pp. 5904–5914

    2021

  32. [32]

    CheXpert: A large chest radiograph dataset with uncertainty labels and expert comparison,

    J. Irvin, P. Rajpurkar, M. Ko, Y . Yu, S. Ciurea-Ilcus, C. Chute, H. Mark- lund, B. Haghgoo, R. Ball, K. Shpanskaya, J. Seekins, D. A. Mong, S. S. Halabi, J. K. Sandberg, R. Jones, D. B. Larson, C. P. Langlotz, B. N. Patel, M. P. Lungren, and A. Y . Ng, “CheXpert: A large chest radiograph dataset with uncertainty labels and expert comparison,” inProceedin...

    2019

  33. [33]

    The earth mover’s distance as a metric for image retrieval,

    Y . Rubner, C. Tomasi, and L. J. Guibas, “The earth mover’s distance as a metric for image retrieval,”International Journal of Computer Vision, vol. 40, no. 2, pp. 99–121, 2000

    2000

  34. [34]

    Villani,Optimal Transport: Old and New

    C. Villani,Optimal Transport: Old and New. Springer, 2009

    2009

  35. [35]

    Santambrogio,Optimal Transport for Applied Mathematicians

    F. Santambrogio,Optimal Transport for Applied Mathematicians. Birkh¨auser, 2015

    2015

  36. [36]

    Computational optimal transport,

    G. Peyr ´e and M. Cuturi, “Computational optimal transport,”Foundations and Trends in Machine Learning, vol. 11, no. 5–6, pp. 355–607, 2019

    2019

  37. [37]

    Sinkhorn distances: Lightspeed computation of optimal transport,

    M. Cuturi, “Sinkhorn distances: Lightspeed computation of optimal transport,” inAdvances in Neural Information Processing Systems, vol. 26, 2013, pp. 2292–2300

    2013

  38. [38]

    Near-linear time approximation algorithms for optimal transport via sinkhorn iteration,

    J. Altschuler, J. Weed, and P. Rigollet, “Near-linear time approximation algorithms for optimal transport via sinkhorn iteration,” inAdvances in Neural Information Processing Systems, vol. 30, 2017

    2017

  39. [39]

    From word embeddings to document distances,

    M. J. Kusner, Y . Sun, N. I. Kolkin, and K. Q. Weinberger, “From word embeddings to document distances,” inProceedings of the 32nd International Conference on Machine Learning, vol. 37, 2015, pp. 957– 966

    2015

  40. [40]

    The optimal partial transport problem,

    A. Figalli, “The optimal partial transport problem,”Archive for Rational Mechanics and Analysis, vol. 195, no. 2, pp. 533–560, 2010

    2010

  41. [41]

    Scaling algo- rithms for unbalanced transport problems,

    L. Chizat, G. Peyr ´e, B. Schmitzer, and F.-X. Vialard, “Scaling algo- rithms for unbalanced transport problems,”Mathematics of Computation, vol. 87, no. 314, pp. 2563–2609, 2018

    2018

  42. [42]

    MIMIC-CXR, a de- identified publicly available database of chest radiographs with free-text reports,

    A. E. W. Johnson, T. J. Pollard, S. J. Berkowitz, N. R. Greenbaum, M. P. Lungren, C.-y. Deng, R. G. Mark, and S. Horng, “MIMIC-CXR, a de- identified publicly available database of chest radiographs with free-text reports,”Scientific Data, vol. 6, no. 317, 2019

    2019

  43. [43]

    Physiobank, physiotoolkit, and physionet: Components of a new research resource for complex physiologic signals,

    A. L. Goldberger, L. A. N. Amaral, L. Glass, J. M. Hausdorff, P. C. Ivanov, R. G. Mark, J. E. Mietus, G. B. Moody, C.-K. Peng, and H. E. Stanley, “Physiobank, physiotoolkit, and physionet: Components of a new research resource for complex physiologic signals,”Circulation, vol. 101, no. 23, pp. e215–e220, 2000

    2000

  44. [44]

    Preparing a collection of radiology examinations for distribution and retrieval,

    D. Demner-Fushman, M. D. Kohli, M. B. Rosenman, S. E. Shooshan, L. Rodriguez, S. Antani, G. R. Thoma, and C. J. McDonald, “Preparing a collection of radiology examinations for distribution and retrieval,” Journal of the American Medical Informatics Association, vol. 23, no. 2, pp. 304–310, 2016

    2016

  45. [45]

    The Llama 3 Herd of Models

    A. Grattafioriet al., “The Llama 3 herd of models,”arXiv preprint arXiv:2407.21783, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024