Pith. sign in

REVIEW 5 major objections 6 minor 45 references

FairREAD: Re-fusing Demographic Attributes after Disentanglement for Fair Medical Image Classification

T0 review · 5 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read FairREAD re-integrates demographic attributes after disentangling them from chest X-ray representations, achieving the best average AUC and the smallest subgroup gaps among all compared methods on CheXpert.

desk verdict A coherent fairness pipeline with a solid system-level result, but the paper's headline mechanism—disentanglement plus re-fusion—lacks direct evidence; the gains could come mostly from the subgroup-specific thresholds. read the letter →

arxiv 2412.16373 v2 pith:2NIMQORT submitted 2024-12-20 cs.CV cs.AIcs.CY

Yicheng Gao , Jinkui Hao , Bo Zhou This is my paper · ORCID
classification cs.CVcs.AIcs.CY
keywords fairnessmedicalimageclassificationdisentanglementdemographicattributesre-fusionsubgroup-specificthresholdCheXpertchestX-ray
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that removing demographic information from a medical image representation to make a model fair also removes clinically useful information, so a better route is to remove it and then deliberately fuse it back in. The proposed FairREAD framework does exactly that: a fair image encoder trained with orthogonality constraints and an adversarial classifier strips demographic cues, a re-fusion module reintroduces the demographic attributes as scaling parameters, and a subgroup-specific threshold equalizes performance. On the CheXpert chest X-ray benchmark, the paper reports that FairREAD reaches the highest average AUC among all compared methods while posting the smallest average subgroup gaps in AUC and equalized odds, and it is the only method with positive fairness-accuracy trade-off scores on every pathology tested. A sympathetic reader would take the central claim to be that fairness in medical imaging need not come at the cost of diagnostic accuracy.

What carries the argument

The load-bearing mechanism is the FairREAD pipeline with three parts. First, a fair image encoder produces latent vectors $z_T$ that are pushed to be independent of demographic attributes by two orthogonality losses—a column-space loss computed from an SVD of the demographic-attribute latent space and a row-space covariance loss—plus an adversarial classifier that tries to predict age, sex, and race from $z_T$ and is made to fail. Second, a re-fusion block encodes the demographic attribute vector with an MLP into scaling parameters $\mu$ and $\sigma^2$, projects $z_T$ into a lower dimension, rescales it, and multiplies the result back into $z_T$; several re-fusion-plus-convolution blocks follow. Third, a subgroup-specific "Min-gap" threshold selects, for each demographic subgroup, the threshold that minimizes $|\text{TPR}_g - \text{TNR}_g|$.

What would settle it

Train a demographic-attribute classifier on the fair encoder's latent vectors $z_T$: if it predicts age, sex, or race with AUC well above chance, the encoder has not removed demographics, and the re-fusion story collapses. A second check is to run the full pipeline with all subgroup-specific thresholds replaced by one global value; if the reported $\Delta$EO reduction persists, the Min-gap threshold rather than the re-fusion mechanism is doing the fairness work.

Watch

Extended reading notes

Core claim

The central discovery is that re-fusion after disentanglement beats both plain training and earlier fairness methods: on CheXpert, FairREAD achieves an average AUC of 0.839 versus the ERM baseline's 0.819, an average $\Delta$AUC of 0.076, and an average $\Delta$EO of 0.129, the smallest among all compared methods. In out-of-distribution tests on MIMIC-CXR for cardiomegaly and pleural effusion, the same model also outperforms the baselines. The authors attribute this to the architecture: demographic attributes are removed from the learned image representation, then explicitly re-added through a rescaling mechanism, so the classifier can use them without relying on the spurious correlations they carry.

Load-bearing premise

The fair image encoder really has removed demographic information from $z_T$; the only supporting evidence offered is the t-SNE visualization, not a quantitative measure of demographic leakage.

Editorial extensions

If this is right

  • Fairness mitigation does not have to mean discarding demographic information: the paper shows the attributes can be removed from the image representation and then explicitly re-added without restoring the shortcut.
  • Subgroup-specific threshold selection can be bolted onto any baseline, so the Min-gap strategy may become a standard post-processing companion for fair medical classifiers.
  • The out-of-distribution result on MIMIC-CXR suggests FairREAD's gains are not an artifact of the CheXpert training distribution, at least for cardiomegaly and pleural effusion.
  • Because FairREAD is the only compared method with positive fairness-accuracy trade-off scores (FAT EEO, FAT EAUC) on every tested pathology, its advantage over ERM is consistent rather than disease-specific.

Reading between the lines

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

  • The re-fusion module is a generic plug-in: it could be paired with any fair representation learner, such as an adversarial, contrastive, or pruning-based encoder, to test whether the fairness gains come from the encoder or from the re-fusion itself.
  • The subgroup-specific Min-gap threshold directly targets the equalized-odds gap, so a controlled comparison that varies only the threshold would reveal how much of the reported $\Delta$EO improvement is attributable to post-processing rather than representation learning.
  • Because age and race are binarized, the current experiments cannot tell whether re-fusion amplifies or relieves disparities inside coarse groups; evaluating on finer-grained or intersectional subgroups is the natural next test.
  • If a quantitative leakage probe ever shows demographic information surviving in $z_T$, the method's stated mechanism would reduce to thresholding plus a possibly leaky encoder, so measuring leakage should precede clinical translation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes FairREAD, a framework for fair medical image classification consisting of three components: a fair image encoder trained with column- and row-space orthogonality losses plus adversarial training to remove demographic information from the latent representation; a re-fusion module that re-injects encoded demographic attributes through feature rescaling; and a subgroup-specific Min-gap threshold that balances TPR and TNR within each subgroup. The method is evaluated on CheXpert for four pathologies with three binarized demographic attributes, and an out-of-distribution transfer to MIMIC-CXR, against five baselines (ERM, FCRO, LNTL, FairAdaBN, AL). Table 2 shows that FairREAD achieves the best average AUC and the lowest average ΔAUC and ΔEO, with the best FAT EEO and FAT EAUC on average.

Significance. The paper addresses an important and practical problem: sensitive attributes in medical imaging are often clinically informative, so methods that remove them entirely can degrade diagnostic performance. The idea of explicitly re-fusing encoded demographic attributes after disentanglement is a reasonable approach to manage this trade-off, and the comparison is made fairer than in many prior works by applying the same subgroup-specific thresholding protocol to all baselines. The inclusion of an OOD experiment on MIMIC-CXR and the release of code are additional strengths. However, the current evidence does not yet substantiate the central attribution claim that the disentanglement stage is what drives the observed gains, and the reported average accuracy is lower than ERM, so the abstract's claim of maintaining diagnostic accuracy is not supported. These are fixable with additional ablations and clarifications.

major comments (5)
  1. [Section 2.1, Eqs. (1)-(3)] The fair image encoder is the core novelty, but its implementation is under-specified and its output is not quantitatively validated. The SVD truncation rank k in Eq. (1) is never stated; Eq. (3) is dimensionally unclear because ^Z^i_T is defined as a row vector for one sample while the double sum runs over i,j = 1..d_z, and the meaning of 'row-wise mean vector' for a matrix whose rows are samples is ambiguous. Furthermore, the only evidence that zT is invariant to demographic attributes is the t-SNE visualization in Fig. 4, which is not a quantitative leakage measure. Please specify k and the exact computation of Eq. (3), and report the AUC of a classifier trained to predict each demographic attribute from zT, or an equivalent information-leakage metric.
  2. [Section 3.1, Table 2] The abstract's claim of 'maintaining diagnostic accuracy' is contradicted by the reported average accuracy in Table 2: FairREAD obtains 0.736±0.010, below ERM's 0.744±0.007 and well below FCRO's 0.782±0.010. The claim is supported only for AUC (0.839 vs 0.819). Please either soften the accuracy claim or provide a statistical test (e.g., paired test across folds) demonstrating that the accuracy differences are not significant. In addition, the FAT EEO and FAT EAUC columns in Tables 2 and 3, and all entries in Table 4, lack standard deviations, which is inconsistent with the paper's emphasis on stability across folds.
  3. [Section 2.3, Table 3] The Min-gap threshold in Eq. (10) is a post-processing step that contributes substantially to the reported fairness improvement: in the ablation, ΔEO drops from 0.377 (default threshold) to 0.129 (Min-gap). Although the same thresholding is applied to all baselines (Section 2.6), the paper's central claim that 're-fusing after disentanglement' is the driver of the gains requires an ablation that keeps the re-fusion module and Min-gap threshold fixed while replacing the fair image encoder with an ERM encoder. The αadv ablation in Fig. 3 does not serve this purpose because the stage-1 orthogonality losses remain active even when αadv=0. Please add this ablation, or explicitly re-scope the claim to the full pipeline rather than to the disentanglement step.
  4. [Section 2.7, Eq. (19)] The definition of ΔED (equal-odds disparity) is written incorrectly: the formula as printed is max_{y,a0,a1} |P(Ŷ=0|Y=y,A=a0) − P(Ŷ=y|Y=0,A=a1)|, which conditions the second term on Y=0 while the first is conditioned on Y=y, and the predicted events also do not match. This definition cannot be evaluated. Please replace it with the standard equal-odds formula, e.g., max_{y∈{0,1}, a0,a1∈A} |P(Ŷ=y|Y=y,A=a0) − P(Ŷ=y|Y=y,A=a1)|, and explicitly define ΔEO, which is used in Eqs. (20)-(21) but never defined.
  5. [Section 2.2, Eq. (9)] The re-fusion operation is not fully specified: the demographic MLP is said to have 'two outputs' (μ and σ²) without stating their dimensions, and Eq. (9) does not indicate how the element-wise product with zT is broadcast when Proj maps to a lower dimension dhidden. Please specify μ, σ² ∈ R^{dhidden} (or equivalent) and clarify the tensor shapes in Eq. (9), as this is needed for reproduction.
minor comments (6)
  1. [Section 1, Introduction] The phrase 'a re-fusion mechanism that that incorporates' contains a duplicated 'that'.
  2. [Section 2.1, Eqs. (4)-(5)] The adversarial loss formula is written with an ambiguous double negative; Eq. (5) should be rewritten with parentheses, e.g., Ladv = -Σ_i[-a_i log â_i + (1-a_i) log(1-â_i)].
  3. [Section 2.5] The subsampling procedure used to magnify subgroup disparity (e.g., which subgroups are downsampled and at what rate) is not described; please add this detail.
  4. [Section 3.3] The t-SNE of the mean and std outputs of the attribute MLP shows separation because these are direct functions of the attribute vector a; this does not provide evidence about the re-fusion mechanism and should be labeled accordingly.
  5. [Appendix A, Table A.6] The caption says 'final models on MIMIC' but the models are trained on CheXpert; the caption should say 'final models used for CheXpert evaluation and MIMIC-CXR transfer' or similar.
  6. [Abstract] The claim of establishing 'a new benchmark' overstates the paper's contribution; the paper introduces a method and an empirical comparison, not a benchmark.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: FairREAD's components are independently evaluated, and the Min-gap threshold is a transparent post-processing step applied equally to all baselines.

full rationale

I examined the derivation chain from the fair image encoder (Sec. 2.1) through re-fusion (Sec. 2.2) and Min-gap thresholding (Sec. 2.3) to the reported metrics (Tables 2 and 3). The threshold rule (Eq. 10) is fit on the training set and then applied to held-out test predictions; the paper explicitly states that the same subgroup-specific threshold selection is applied to every baseline (Sec. 2.6), so the comparison is not biased by the post-processing. Table 3 is an honest ablation of threshold-selection strategies on a fixed model, not a circular re-statement of the objective. The remaining weaknesses—the lack of a quantitative leakage metric for zT and the absence of an ERM-encoder control for the re-fusion module—are evidential gaps in attribution, not instances where a claimed result is equal to its input by construction. No load-bearing self-citation or imported-uniqueness argument is present; the orthogonality losses are attributed to Deng et al. (2023), an external source, and hyperparameters are chosen by grid search on validation folds. I therefore find no circular step under the definitions in the task.

Assumptions & free parameters 9 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a set of hand-selected hyperparameters, an unverified disentanglement assumption, and a threshold-fitting rule that directly targets the reported fairness metric. No new physical entities are introduced.

free parameters (9)
  • alpha_adv = 0.1 (Cardiomegaly, Pleural Effusion), 0.5 (Pneumonia, Fracture)
    Adversarial loss weight in Eq. (17); selected by grid search on validation performance (Table A.6).
  • Number of re-fusion blocks N = 2 (Cardiomegaly, Pleural Effusion), 1 (Pneumonia, Fracture)
    Grid-searched over {1,2,3}.
  • Hidden dimension d_hidden = 256
    Grid-searched over {256, 1024}.
  • Dropout rate = 0.1
    Grid-searched over {0.1, 0.3}.
  • SVD truncation k = not specified
    Rank of column space in Eq. (1)-(2); no value is given anywhere in the paper.
  • Subgroup-specific thresholds theta_g = 8 thresholds per target (fit on training set)
    Min-gap strategy, Eq. (10): directly minimizes |TPR-TNR| per subgroup, targeting Delta-EO.
  • Orthogonality weights lambda_c, lambda_r = taken from Deng et al. 2023
    Borrowed, not re-fit; stated in Appendix A.
  • Dataset subsampling disparity = 16% for Cardiomegaly/Pleural Effusion, 10% for Pneumonia/Fracture
    Hand-chosen to magnify subgroup disparity; Section 2.5.
  • Age cutoff and race binarization = 60 years; White vs Non-white
    Discretization choices; Section 2.5.
assumptions (4)
  • domain assumption CheXpert labels and demographic metadata are correct ground truth
    All metrics are computed against these labels; no label-noise analysis is provided.
  • domain assumption Orthogonality and adversarial losses yield a demographically invariant zT
    The paper asserts this and shows t-SNE (Figure 4), but never measures demographic-attribute leakage quantitatively from zT.
  • domain assumption Delta-EO and Delta-AUC are the operative definitions of fairness
    The Min-gap threshold and the evaluation both optimize/report these metrics; other fairness notions are not considered.
  • domain assumption MIMIC-CXR is a valid out-of-distribution test for CheXpert-trained models
    Section 3.4 uses it as OOD without demonstrating label/annotation compatibility for the two transferred pathologies.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FairREAD: Re-fusing Demographic Attributes after Disentanglement for Fair Medical Image Classification." pith.science (2026). https://pith.science/paper/2NIMQORT

@misc{pith2026241216373,
  author       = {Pith},
  title        = {Pith review of: FairREAD: Re-fusing Demographic Attributes after Disentanglement for Fair Medical Image Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2NIMQORT}},
  note         = {Machine review of arXiv:2412.16373}
}
read the original abstract

Recent advancements in deep learning have shown transformative potential in medical imaging, yet concerns about fairness persist due to performance disparities across demographic subgroups. Existing methods aim to address these biases by mitigating sensitive attributes in image data; however, these attributes often carry clinically relevant information, and their removal can compromise model performance-a highly undesirable outcome. To address this challenge, we propose Fair Re-fusion After Disentanglement (FairREAD), a novel, simple, and efficient framework that mitigates unfairness by re-integrating sensitive demographic attributes into fair image representations. FairREAD employs orthogonality constraints and adversarial training to disentangle demographic information while using a controlled re-fusion mechanism to preserve clinically relevant details. Additionally, subgroup-specific threshold adjustments ensure equitable performance across demographic groups. Comprehensive evaluations on a large-scale clinical X-ray dataset demonstrate that FairREAD significantly reduces unfairness metrics while maintaining diagnostic accuracy, establishing a new benchmark for fairness and performance in medical image classification.

Figures

Figures reproduced from arXiv: 2412.16373 by the authors.

Figure 1
Figure 1. Model architecture of FairREAD. The input chest X-ray image is first encoded by the Fair Image Encoder (FIE) into a demographic [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Distribution of demographic subgroups in the processed CheXpert dataset. “NW” and “W” denotes “Non-White” and “White”; “M” and [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Performance of FairREAD on classifying patients with Fracture using different values of αadv. Optimal value of αadv in terms of FAT EEO is marked in purple dashed lines in the figures for Accuracy, ∆EO, and FAT EEO. Optimal value of αadv in terms of FAT EAUC is marked in orange dashed lines in the figures for AUC, ∆AUC, and FAT EAUC. The value in each plot when αadv = 0 is marked with a blue star. (a) Embedding spac… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: t-SNE visualization of embedding space in FairREAD model for Cardiomegaly classification (best viewed in color). “ERM” represents [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 32 canonical work pages

  1. [1]

    , author Demir, A

    author Ashames, M.M.A. , author Demir, A. , author Gerek, O.N. , author Fidan, M. , author Gulmezoglu, M.B. , author Ergin, S. , author Edizkan, R. , author Koc, M. , author Barkana, A. , author Calisir, C. , year 2024 . title Are deep learning classification results obtained on ct scans fair and interpretable? journal Physical and Engineering Sciences in...

    work page doi:10.1007/s13246-024-01419-8 2024

  2. [2]

    shortcuts

    author Banerjee, I. , author Bhattacharjee, K. , author Burns, J.L. , author Trivedi, H. , author Purkayastha, S. , author Seyyed-Kalantari, L. , author Patel, B.N. , author Shiradkar, R. , author Gichoya, J. , year 2023 . title “shortcuts” causing bias in radiology artificial intelligence: causes, evaluation and mitigation. journal Journal of the America...

    work page 2023

  3. [3]

    , author Atapour-Abarghouei, A

    author Bevan, P.J. , author Atapour-Abarghouei, A. , year 2022 a. title Detecting melanoma fairly: Skin tone detection and debiasing for skin lesion classification , in: booktitle MICCAI Workshop on Domain Adaptation and Representation Transfer , organization Springer . pp. pages 1--11

    work page 2022

  4. [4]

    , author Atapour-Abarghouei, A

    author Bevan, P.J. , author Atapour-Abarghouei, A. , year 2022 b. title Detecting melanoma fairly: Skin tone detection and debiasing for skin lesion classification , in: booktitle DART@MICCAI . https://api.semanticscholar.org/CorpusID:246634884

    work page 2022

  5. [5]

    , author Gebru, T

    author Buolamwini, J. , author Gebru, T. , year 2018 . title Gender shades: Intersectional accuracy disparities in commercial gender classification , in: editor Friedler, S.A. , editor Wilson, C. (Eds.), booktitle Proceedings of the 1st Conference on Fairness, Accountability and Transparency , publisher PMLR . pp. pages 77--91 . https://proceedings.mlr.pr...

    work page 2018

  6. [6]

    , author Wang, J.J

    author Chen, R.J. , author Wang, J.J. , author Williamson, D.F.K. , author Chen, T.Y. , author Lipkova, J. , author Lu, M.Y. , author Sahai, S. , author Mahmood, F. , year 2023 . title Algorithmic fairness in artificial intelligence for medicine and healthcare . journal Nature Biomedical Engineering volume 7 , pages 719--742 . https://doi.org/10.1038/s415...

    doi:10.1038/s41551-023-01056-8 2023

  7. [7]

    , author Zhan, C

    author Deho, O.B. , author Zhan, C. , author Li, J. , author Liu, J. , author Liu, L. , author Duy Le, T. , year 2022 . title How do the existing fairness metrics and unfairness mitigation algorithms contribute to ethical learning analytics? journal British Journal of Educational Technology volume 53 , pages 822--843 . :https://doi.org/10.1111/bjet.13217

    work page doi:10.1111/bjet.13217 2022

  8. [8]

    On Fairness of Medical Image Classification with Multiple Sensitive Attributes via Learning Orthogonal Representations

    author Deng, W. , author Zhong, Y. , author Dou, Q. , author Li, X. , year 2023 . title On fairness of medical image classification with multiple sensitive attributes via learning orthogonal representations . https://arxiv.org/abs/2301.01481, arXiv:2301.01481 http://arxiv.org/abs/2301.01481

    work page Pith review arXiv 2023

Show all 45 references
  1. [9]

    , author Hers, B

    author Du, S. , author Hers, B. , author Bayasi, N. , author Hamarneh, G. , author Garbi, R. , year 2022 . title Fairdisco: Fairer ai in dermatology via disentanglement contrastive learning . https://arxiv.org/abs/2208.10013, arXiv:2208.10013 http://arxiv.org/abs/2208.10013

    2022 arXiv

  2. [10]

    , author Xue, Y

    author Du, Y. , author Xue, Y. , author Dharmakumar, R. , author Tsaftaris, S.A. , year 2023 . title Unveiling fairness biases in deep learning-based brain mri reconstruction . https://arxiv.org/abs/2309.14392, arXiv:2309.14392 http://arxiv.org/abs/2309.14392

    2023 arXiv

  3. [11]

    , author Hardt, M

    author Dwork, C. , author Hardt, M. , author Pitassi, T. , author Reingold, O. , author Zemel, R. , year 2011 . title Fairness through awareness . https://arxiv.org/abs/1104.3913, arXiv:1104.3913 http://arxiv.org/abs/1104.3913

    2011 arXiv

  4. [12]

    title Artificial intelligence and machine learning (ai/ml)-enabled medical devices

    author FDA , year 2024 . title Artificial intelligence and machine learning (ai/ml)-enabled medical devices . https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-aiml-enabled-medical-devices

    2024

  5. [13]

    , author Banerjee, I

    author Gichoya, J.W. , author Banerjee, I. , author Bhimireddy, A.R. , author Burns, J.L. , author Celi, L.A. , author Chen, L.C. , author Correa, R. , author Dullerud, N. , author Ghassemi, M. , author Huang, S.C. , author Kuo, P.C. , author Lungren, M.P. , author Palmer, L.J...

    2022

  6. [14]

    , author Jones, C

    author Glocker, B. , author Jones, C. , author Bernhardt, M. , author Winzeck, S. , year 2023 . title Algorithmic encoding of protected characteristics in chest x-ray disease detection models . journal eBioMedicine volume 89 . https://doi.org/10.1016/j.ebiom.2023.104467, :10.1...

    2023

  7. [15]

    , author Price, E

    author Hardt, M. , author Price, E. , author Srebro, N. , year 2016 . title Equality of opportunity in supervised learning . journal Advances in neural information processing systems volume 29

    2016

  8. [16]

    , author Arjovsky, M

    author Idrissi, B.Y. , author Arjovsky, M. , author Pezeshki, M. , author Lopez-Paz, D. , year 2022 . title Simple data balancing achieves competitive worst-group-accuracy . https://arxiv.org/abs/2110.14503, arXiv:2110.14503 http://arxiv.org/abs/2110.14503

    2022 arXiv

  9. [17]

    , author Rajpurkar, P

    author Irvin, J. , author Rajpurkar, P. , author Ko, M. , author Yu, Y. , author Ciurea-Ilcus, S. , author Chute, C. , author Marklund, H. , author Haghgoo, B. , author Ball, R. , author Shpanskaya, K. , author Seekins, J. , author Mong, D.A. , author Halabi, S.S. , author San...

    2019 arXiv

  10. [18]

    , author Nachum, O

    author Jiang, H. , author Nachum, O. , year 2020 . title Identifying and correcting label bias in machine learning , in: booktitle International conference on artificial intelligence and statistics , organization PMLR . pp. pages 702--712

    2020

  11. [19]

    , author Xu, Z

    author Jin, R. , author Xu, Z. , author Zhong, Y. , author Yao, Q. , author Dou, Q. , author Zhou, S.K. , author Li, X. , year 2024 . title Fairmedfm: Fairness benchmarking for medical imaging foundation models . https://arxiv.org/abs/2407.00983, arXiv:2407.00983 http://arxiv....

    2024 arXiv

  12. [20]

    , author Lum, K

    author Johndrow, J.E. , author Lum, K. , year 2019 . title An algorithm for removing sensitive information . journal The Annals of Applied Statistics volume 13 , pages 189--220

    2019

  13. [21]

    , author Pollard, T.J

    author Johnson, A.E.W. , author Pollard, T.J. , author Greenbaum, N.R. , author Lungren, M.P. , author ying Deng, C. , author Peng, Y. , author Lu, Z. , author Mark, R.G. , author Berkowitz, S.J. , author Horng, S. , year 2019 . title Mimic-cxr-jpg, a large publicly available ...

    2019 arXiv

  14. [22]

    , author Jang, T

    author Jung, H. , author Jang, T. , author Wang, X. , year 2024 . title A unified debiasing approach for vision-language models across modalities and tasks . https://arxiv.org/abs/2410.07593, arXiv:2410.07593 http://arxiv.org/abs/2410.07593

    2024 arXiv

  15. [23]

    , author Kim, H

    author Kim, B. , author Kim, H. , author Kim, K. , author Kim, S. , author Kim, J. , year 2019 . title Learning not to learn: Training deep neural networks with biased data . https://arxiv.org/abs/1812.10352, arXiv:1812.10352 http://arxiv.org/abs/1812.10352

    2019 arXiv

  16. [24]

    , author Nieto, N

    author Larrazabal, A.J. , author Nieto, N. , author Peterson, V. , author Milone, D.H. , author Ferrante, E. , year 2020 . title Gender imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis . journal Proceedings of the National Academy ...

    2020

  17. [25]

    , author Oliver, N

    author Lepri, B. , author Oliver, N. , author Letouz \'e , E. , author Pentland, A. , author Vinck, P. , year 2018 . title Fair, transparent, and accountable algorithmic decision-making processes: The premise, the proposed solutions, and the open challenges . journal Philosoph...

    2018

  18. [26]

    , author Shi, X

    author Li, B. , author Shi, X. , author Gao, H. , author Jiang, X. , author Zhang, K. , author Harmanci, A.O. , author Malin, B. , year 2023 . title Enhancing fairness in disease prediction by optimizing multiple domain adversarial networks . journal BioRxiv

    2023

  19. [27]

    , author Cui, Z

    author Li, X. , author Cui, Z. , author Wu, Y. , author Gu, L. , author Harada, T. , year 2021 . title Estimating and improving fairness with adversarial learning . https://arxiv.org/abs/2103.04243, arXiv:2103.04243 http://arxiv.org/abs/2103.04243

    2021 arXiv

  20. [28]

    , author Hinton, G

    author van der Maaten, L. , author Hinton, G. , year 2008 . title Visualizing data using t-sne . journal Journal of Machine Learning Research volume 9 , pages 2579--2605 . http://jmlr.org/papers/v9/vandermaaten08a.html

    2008

  21. [29]

    , author Eickhoff, S.B

    author More, S. , author Eickhoff, S.B. , author Caspers, J. , author Patil, K.R. , year 2021 . title Confound removal and normalization in practice: A neuroimaging based sex prediction case study , in: booktitle Machine Learning and Knowledge Discovery in Databases. Applied D...

    2021

  22. [30]

    , author Bakker, M.A

    author Noriega-Campero, A. , author Bakker, M.A. , author Garcia-Bulle, B. , author Pentland, A. , year 2019 . title Active fairness in algorithmic decision making , in: booktitle Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society , pp. pages 77--83

    2019

  23. [31]

    , author Hadzic, A

    author Paul, W. , author Hadzic, A. , author Joshi, N. , author Alajaji, F. , author Burlina, P. , year 2022 . title Tara: Training and representation alteration for ai fairness and domain generalization . journal Neural Computation volume 34 , pages 716--753

    2022

  24. [32]

    , author Ruijsink, B

    author Puyol-Ant \'o n, E. , author Ruijsink, B. , author Piechnik, S.K. , author Neubauer, S. , author Petersen, S.E. , author Razavi, R. , author King, A.P. , year 2021 . title Fairness in cardiac mr image analysis: An investigation of bias due to data imbalance in deep lear...

    2021

  25. [33]

    , author Rodriguez, L

    author Salimi, B. , author Rodriguez, L. , author Howe, B. , author Suciu, D. , year 2019 . title Interventional fairness: Causal database repair for algorithmic fairness , in: booktitle Proceedings of the 2019 International Conference on Management of Data , pp. pages 793--810

    2019

  26. [34]

    , author Navab, N

    author Sarhan, M.H. , author Navab, N. , author Eslami, A. , author Albarqouni, S. , year 2020 . title Fairness by learning orthogonal disentangled representations . https://arxiv.org/abs/2003.05707, arXiv:2003.05707 http://arxiv.org/abs/2003.05707

    2020 arXiv

  27. [35]

    , author Liu, G

    author Seyyed-Kalantari, L. , author Liu, G. , author McDermott, M. , author Chen, I.Y. , author Ghassemi, M. , year 2021 . title CheXclusion : Fairness gaps in deep chest x-ray classifiers . journal Pac Symp Biocomput volume 26 , pages 232--243

    2021

  28. [36]

    , author Zhao, Q

    author Vento, A. , author Zhao, Q. , author Paul, R. , author Pohl, K.M. , author Adeli, E. , year 2022 . title A penalty approach for normalizing feature distributions to build confounder-free models , in: booktitle International Conference on Medical Image Computing and Comp...

    2022

  29. [37]

    , author Vera, F

    author Wadsworth, C. , author Vera, F. , author Piech, C. , year 2018 . title Achieving fairness through adversarial learning: an application to recidivism prediction . https://arxiv.org/abs/1807.00199, arXiv:1807.00199 http://arxiv.org/abs/1807.00199

    2018 arXiv

  30. [38]

    , author Zha, D

    author Wan, M. , author Zha, D. , author Liu, N. , author Zou, N. , year 2023 . title In-processing modeling techniques for machine learning fairness: A survey . journal ACM Transactions on Knowledge Discovery from Data volume 17 , pages 1--27

    2023

  31. [39]

    , author Zeng, D

    author Wu, Y. , author Zeng, D. , author Xu, X. , author Shi, Y. , author Hu, J. , year 2022 . title Fairprune: Achieving fairness through pruning for dermatological disease diagnosis . https://arxiv.org/abs/2203.02110, arXiv:2203.02110 http://arxiv.org/abs/2203.02110

    2022 arXiv

  32. [40]

    , author Li, J

    author Xu, Z. , author Li, J. , author Yao, Q. , author Li, H. , author Zhao, M. , author Zhou, S.K. , year 2024 . title Addressing fairness issues in deep learning-based medical image analysis: a systematic review . journal npj Digital Medicine volume 7 , pages 286

    2024

  33. [41]

    , author Zhao, S

    author Xu, Z. , author Zhao, S. , author Quan, Q. , author Yao, Q. , author Zhou, S.K. , year 2023 . title Fairadabn: Mitigating unfairness with adaptive batch normalization and its application to dermatological disease classification . https://arxiv.org/abs/2303.08325, arXiv:...

    2023 arXiv

  34. [43]

    , author Zhang, H

    author Yang, Y. , author Zhang, H. , author Gichoya, J.W. , author Katabi, D. , author Ghassemi, M. , year 2024 b. title The limits of fair medical imaging ai in real-world generalization . journal Nature Medicine volume 30 , pages 2838--2848 . https://doi.org/10.1038/s41591-0...

    2024 doi

  35. [44]

    , author Dullerud, N

    author Zhang, H. , author Dullerud, N. , author Roth, K. , author Oakden-Rayner, L. , author Pfohl, S.R. , author Ghassemi, M. , year 2022 . title Improving the fairness of chest x-ray classifiers . https://arxiv.org/abs/2203.12609, arXiv:2203.12609 http://arxiv.org/abs/2203.12609

    2022 arXiv

  36. [45]

    , author Adeli, E

    author Zhao, Q. , author Adeli, E. , author Pohl, K.M. , year 2020 . title Training confounder-free deep learning models for medical applications . journal Nature communications volume 11 , pages 6010

    2020

  37. [46]

    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 11, 2026 · model on record in the stance chip above.