Pith. sign in

REVIEW 6 major objections 5 minor 68 references

Fair Distillation: Teaching Fairness from Biased Teachers in Medical Imaging

T0 review · 6 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read FairDi, a teacher-student fairness method, claims to lift overall accuracy and shrink group gaps at the same time in medical imaging.

desk verdict FairDi is a plausible distillation-based fairness method whose broad experiments are undermined by missing variance and inconsistent baseline numbers; the idea deserves review but the central claim is not yet supported. read the letter →

arxiv 2411.11939 v1 pith:GGMYXQ23 submitted 2024-11-18 cs.CV

classification cs.CV
keywords FairDifairnessinmedicalimagingknowledgedistillationgroupAUCgapteacher-studentlearningimageclassificationsegmentation
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 introduces FairDi, a fairness method for medical image classification and segmentation that separates three goals that usually fight each other: overall accuracy, accuracy for each demographic group, and small accuracy differences between groups. FairDi first trains a shared backbone with a fairness-aware loss, then fine-tunes a separate 'teacher' classification head for each demographic group to maximize that group's accuracy, and finally trains one 'student' head that distills the teachers' softened predictions while also being trained for overall accuracy and fairness. On five chest, skin, and eye classification benchmarks and one segmentation benchmark, the paper reports that FairDi achieves the best average overall, worst-group, and equity-scaled AUC, and the smallest average AUC gap, among the compared methods. The significance, if the results hold, is that fairness does not have to be bought by sacrificing accuracy; the standard trade-off can be shifted by decoupling group-optimized experts from a unified model.

What carries the argument

The machinery is a three-step pipeline built on a frozen feature extractor $f_{\theta^*}$ trained with the Fair Identity Scaling (FIS) loss, a loss that reweights each sample by individual difficulty and by the optimal-transport distance between a group's loss distribution and the overall loss distribution. Step 1 adds a randomly initialized classification head $h_{\phi_g^{(t)}}$ for each demographic group $g$ and fine-tunes only that head on the group's samples to maximize group-specific accuracy. Step 2 adds a student head $h_{\phi^{(s)}}$ and minimizes $\lambda\tau^2\,\mathrm{KL}\big(h_{\phi^{(s)}}(f_{\theta^*}(x))\,\|\,h_{\phi_a^{(t)},\tau}(f_{\theta^*}(x))\big) + (1-\lambda)\,w(x,y,a,c)\,\ell\big(h_{\phi^{(s)}}(f_{\theta^*}(x)), y\big)$ with $\lambda=0.95$ and $\tau=1.5$. The per-sample selection of teacher $a$ is what lets the student inherit each group's knowledge; the FIS term is what keeps the student from drifting toward any single teacher and anchors overall accuracy and fairness.

What would settle it

Train FairDi on a synthetic two-group classification problem in which the optimal group-specific decision boundaries on the frozen feature space provably cross (for example, mirrored class-conditional Gaussians per group). If the student's group-specific AUCs fall substantially below the teachers' group-specific AUCs while the AUC gap remains above the teacher gap, the single-shared-head assumption is refuted; a null result would support the paper's claim that the shared head can carry all group-specific functions.

Watch

Extended reading notes

Core claim

The central claim is that a teacher-student decomposition changes the fairness-accuracy Pareto front in medical imaging. Instead of one model trying to satisfy all objectives at once, FairDi trains a teacher head per demographic cohort on a frozen backbone, each greedily optimized only for its own group's classification accuracy, then trains a single student head that, for every sample, matches the softened output of that sample's group teacher through a KL divergence term while also minimizing a sample-weighted classification loss. The paper reports that this yields the best average overall AUC (0.9137), best average minimum subgroup AUC (0.9050), best average equity-scaled AUC (0.9062), and smallest average AUC gap (0.0166) across HAM10000, Fitzpatrick17k, PAPILA, CheXpert, and MIMIC-CXR, and similarly improved equity-scaled Dice for optic-cup and optic-rim segmentation, relative to the compared fairness methods. The claim is that FairDi is the first fairness method to decompose group-accuracy maximization from overall-accuracy maximization and inter-group gap minimization.

Load-bearing premise

The student is a single linear classification head on the same frozen features as the teachers, so the method stands or falls on whether one shared head can approximate every group-specific teacher's outputs for every sample; if the features force the per-group classifiers to disagree too strongly, the distillation term averages conflicting targets and the student loses both accuracy and fairness.

Editorial extensions

If this is right

  • FairDi reports the best average overall AUC (0.9137) and best average minimum-group AUC (0.9050) across the five classification benchmarks, so a practitioner wanting both accuracy and worst-group protection would get both from one training scheme.
  • The average AUC gap drops to 0.0166, below the best previously reported gap, meaning the margin between the strongest and weakest demographic cohort nearly disappears on average.
  • On the FairSeg segmentation benchmark, FairDi reports the best average overall Dice (0.8260) and equity-scaled Dice (0.8062), so the decomposition transfers from binary classification to dense pixel prediction.
  • Since only the small classification heads are trainable in Steps 1 and 2, FairDi can be applied on top of an existing frozen medical imaging backbone with total training time around 40 minutes on HAM10000 and no added test-time cost.

Reading between the lines

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

  • The paper does not test intersectional groups, but nothing in the method forbids defining a teacher for each intersection of sensitive attributes; however, the single-shared-head assumption becomes harder to satisfy as the number of cohorts grows, so the method's advantage may shrink with more groups.
  • A direct probe of the capacity assumption would be to report student versus teacher per-group AUCs on the same held-out data; if the student closely tracks its teachers, the distillation is faithfully transferring knowledge, and if not, the gap between them quantifies the cost of the shared head.
  • The frozen-backbone design suggests FairDi could be combined with any representation learning scheme, not just the FIS-trained backbone, although the fairness of the final student would then depend on whether the frozen features preserve enough group-specific information for the teachers to excel.
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

6 major / 5 minor

Summary. The paper introduces Fair Distillation (FairDi), a two-stage knowledge-distillation method for fairness in medical imaging classification and segmentation. Step 0 trains a backbone with the FIS loss; Step 1 fine-tunes one linear classification head per demographic group on the frozen backbone to create biased group-specific teachers; Step 2 trains a single student head with a KL distillation loss to the per-group teachers combined with the FIS loss. The authors claim that FairDi is the first fairness method to decompose group-accuracy maximization from overall-accuracy and inter-group-gap optimization, and report improved overall AUC, minimum-group AUC, ES-AUC, and reduced AUC Gap relative to ERM, GroupDRO, SWAD, and FIS on five classification datasets, plus improved Dice/IoU and equity-scaled metrics on the Harvard-FairSeg segmentation dataset.

Significance. The proposed decomposition is conceptually clean and practically attractive: if the reported gains hold, FairDi would offer a simple and adaptable way to improve both accuracy and fairness by using biased teachers, and its application to both classification and segmentation gives it broad potential impact. The paper also has strengths: the method is clearly specified, the ablation study isolates the contributions of CutMix, FIS, and the teacher-student steps, and the authors include detailed supplementary tables and fairness metric definitions. However, the central empirical claim currently rests on single-split, single-run results with no variance information, and there are internal inconsistencies between the main tables and the supplementary table. These problems are load-bearing because the paper's contribution is primarily empirical superiority over prior methods.

major comments (6)
  1. [§4.4, Table 3 vs. §9, Table 11] The same baseline is reported with different numbers across tables: ERM HAM10000 Gender Overall AUC is 0.8890 in Table 3 but 0.8520 in Table 11, and PAPILA Gender ERM Overall AUC is 0.8433 in Table 3 but 0.7840 in Table 11. The manuscript does not explain whether these are different splits, different seeds, or different preprocessing choices. Without a clear protocol, the single-split averages in Table 1 are not reproducible and the cross-method comparison is undermined.
  2. [§4.2 and Figure 3] All reported classification results appear to be single runs on a single split; no seeds, confidence intervals, or error bars are reported. The Friedman/Nemenyi tests in Section 4.2 and Figure 3 use one observation per algorithm per dataset, so the statement that FairDi is 'significantly better than all others' for worst-case AUC is not statistically supported. Please provide repeated-seed results with variance estimates and base significance claims on distributions over runs, or state explicitly that the CD diagrams are illustrative only.
  3. [§4.1 and Table 3] The paper states that CutMix is used for the backbone, teacher, and student models, but it does not state whether the ERM, GroupDRO, SWAD, and FIS baselines also use CutMix. The ablation in Section 4.4 shows that adding CutMix changes ERM PAPILA Gender Overall AUC from 0.8433 to 0.9142 and AUC Gap from 0.0180 to 0.1500; if the baselines were trained without CutMix, a substantial part of the reported improvement could be due to augmentation rather than the FairDi mechanism. Please specify the exact training protocol for every baseline.
  4. [Table 1] Table 1 lists identical average rank values for AUC Gap, MeanPSD, and MaxPSD for every method (3.00, 3.55, 3.27, 3.00, 2.18), even though the underlying per-dataset values in Table 11 are clearly different and would not be expected to yield identical ranks across three distinct fairness metrics. This strongly suggests a calculation or transcription error in the rank rows; please recompute and verify all rank statistics.
  5. [§4.2, Table 11] The claim that FairDi 'outperforms existing methods' is stronger than the evidence in Table 11: FIS has a lower AUC Gap than FairDi on MIMIC-CXR Race (0.0103 vs. 0.0160) and CheXpert Age (0.0256 vs. 0.0281), and FairDi does not win every reported cell. The averages in Table 1 hide these exceptions. Please qualify the claim to 'highest average' and provide a per-cell comparison or significance test before asserting global superiority.
  6. [§3.3, Eq. (3)] Equation (3) trains a single linear classification head on the frozen FIS backbone to match the per-group teacher linear heads via KL distillation, yet the manuscript provides no capacity or geometry argument that a single linear map can realize the group-specific teacher functions on the shared features. If the features require different linear maps per group, the per-sample KL term averages conflicting teacher targets and could degrade both accuracy and fairness. Please provide an analysis of this assumption or an experiment varying the student head capacity.
minor comments (5)
  1. [Abstract] Abstract contains 'F air Distillation' with an erroneous space; it should be 'Fair Distillation'.
  2. [§3.3, Eq. (3)] The KL divergence is written as KL[h_student || h_teacher], but the arguments are not explicitly defined as probability distributions over classes; please clarify that h outputs are softmax probabilities and state how the temperature τ enters the KL term.
  3. [§4.1, Datasets] The dataset name is spelled 'Papila' in the datasets list but 'PAPILA' elsewhere; please use consistent spelling.
  4. [Figure 1 caption] The caption says 'Each model's Pareto front includes two points: one maximizing worst-group AUC, the other maximizing overall AUC,' which is unclear because a Pareto front is a set of solutions, not two points; please clarify how the two points are selected and what they represent.
  5. [Supplementary, Eq. (4)] The ES-AUC definition uses AUC(D_g) but D_g is not defined in the supplementary; please add the definition or reference the main text definition from Section 3.2.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the reported gains are empirical and not entailed by the loss definitions or by self-citation.

full rationale

The paper's central claim is that the FairDi student (Eq. 3) improves overall AUC, minimum group AUC, and AUC gap relative to FIS and other baselines on held-out test sets. Nothing in Eqs. (1)-(3) makes these improvements true by construction: the KL term distills per-group teacher predictions, the CE term applies FIS sample weights, and the reported AUC, ES-AUC, and PSD metrics are evaluated on test data rather than being directly optimized. The only notable self-referential element is the use of FIS loss [31] both as Stage-0 pretraining and as a comparison baseline, and [31] shares a co-author with this submission. However, that citation is not load-bearing in a circular way: FIS is an externally published method with its own benchmark results, and the paper's contribution (adding cohort-specific teachers and distillation on top of FIS) is tested independently against FIS on held-out data. Concerns about missing variance estimates, inconsistent baseline numbers across Tables 3 and 11, and the fact that FairDi does not win every per-cell comparison are legitimate experimental-rigor concerns, but they are correctness risks rather than evidence that a result reduces to its inputs by construction.

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

No new physical entities are introduced. The main free parameters are distillation hyperparameters (lambda, tau, c) chosen without sensitivity analysis. The strongest ad hoc assumption is that a single linear student head can imitate all group teachers simultaneously.

free parameters (4)
  • lambda (distillation weight) = 0.95
    Eq. 3 balances KL distillation against FIS loss; chosen by hand, heavily favoring distillation. No ablation over lambda is shown.
  • tau (distillation temperature) = 1.5
    Eq. 3 temperature for teacher logits; set in Section 4.1 without sensitivity analysis.
  • c (FIS balance) = 0.5 for backbone and student, 0 for teachers
    Balance between individual and group scaling in the FIS loss (Eq. 1); used in Steps 0 and 2, chosen without ablation.
  • bias binarization thresholds = age >60, skin type >2
    Sensitive attributes are binarized following MedFair [64]; these thresholds define fairness in the evaluation and are not varied in the paper.
assumptions (4)
  • standard math KL divergence and softmax temperature scaling behave as in standard knowledge distillation (Hinton et al.)
    Invoked in Eq. 3 without proof, standard practice.
  • domain assumption Sensitive attribute labels a_i are available and correct for every training and test sample
    The whole method requires per-sample group labels; the paper relies on the datasets' meta-data.
  • ad hoc to paper A single shared linear student layer can approximate the per-group teacher functions on the frozen FIS features
    Eq. 3 assumes this; no proof or analysis is provided, and it is the main load-bearing premise of Step 2.
  • domain assumption Baselines (ERM, GroupDRO, SWAD, FIS) were trained under comparable conditions, including data augmentation
    The paper does not state whether baselines used CutMix, yet the comparison assumes parity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fair Distillation: Teaching Fairness from Biased Teachers in Medical Imaging." pith.science (2026). https://pith.science/paper/GGMYXQ23

@misc{pith2026241111939,
  author       = {Pith},
  title        = {Pith review of: Fair Distillation: Teaching Fairness from Biased Teachers in Medical Imaging},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GGMYXQ23}},
  note         = {Machine review of arXiv:2411.11939}
}
read the original abstract

Deep learning has achieved remarkable success in image classification and segmentation tasks. However, fairness concerns persist, as models often exhibit biases that disproportionately affect demographic groups defined by sensitive attributes such as race, gender, or age. Existing bias-mitigation techniques, including Subgroup Re-balancing, Adversarial Training, and Domain Generalization, aim to balance accuracy across demographic groups, but often fail to simultaneously improve overall accuracy, group-specific accuracy, and fairness due to conflicts among these interdependent objectives. We propose the Fair Distillation (FairDi) method, a novel fairness approach that decomposes these objectives by leveraging biased ``teacher'' models, each optimized for a specific demographic group. These teacher models then guide the training of a unified ``student'' model, which distills their knowledge to maximize overall and group-specific accuracies, while minimizing inter-group disparities. Experiments on medical imaging datasets show that FairDi achieves significant gains in both overall and group-specific accuracy, along with improved fairness, compared to existing methods. FairDi is adaptable to various medical tasks, such as classification and segmentation, and provides an effective solution for equitable model performance.

Figures

Figures reproduced from arXiv: 2411.11939 by the authors.

Figure 1
Figure 1. Performance comparison of models (ERM [57], GroupDRO [46], SWAD [2], FIS [31], and our FairDi) on the HAM10000 dataset [56] for benign/malignant classification by gender. The left panel shows group-specific AUCs (Male and Fe￾male), and the right panel plots fairness (AUC Gap) vs. Overall AUC. Each model’s Pareto front includes two points: one maxi￾mizing worst-group AUC, the other maximizing overall AUC. Our FairDi … view at source ↗
Figure 2
Figure 2. Diagram of the FairDi training process. After pre [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Performance of fairness algorithms for classification across all datasets as average rank CD diagrams. Our FairDi is the highest [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Performance of fairness segmentation algorithms shown [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 49 canonical work pages

  1. [1]

    Model compression

    Cristian Bucilu ˇa, Rich Caruana, and Alexandru Niculescu- Mizil. Model compression. In Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 535–541, 2006. 3

  2. [2]

    Swad: Domain generalization by seeking flat minima

    Junbum Cha, Sanghyuk Chun, Kyungjae Lee, Han-Cheol Cho, Seunghyun Park, Yunsung Lee, and Sungrae Park. Swad: Domain generalization by seeking flat minima. Advances in Neural Information Processing Systems , 34: 22405–22418, 2021. 1, 2, 6

  3. [3]

    Ethical machine learning in healthcare

    Irene Y Chen, Emma Pierson, Sherri Rose, Shalmali Joshi, Kadija Ferryman, and Marzyeh Ghassemi. Ethical machine learning in healthcare. Annual review of biomedical data science, 4(1):123–144, 2021. 1

  4. [4]

    Transunet: Transformers make strong encoders for medi- cal image segmentation

    Jieneng Chen, Yongyi Lu, Qihang Yu, Xiangde Luo, Ehsan Adeli, Yan Wang, Le Lu, Alan L Yuille, and Yuyin Zhou. Transunet: Transformers make strong encoders for medi- cal image segmentation. arXiv preprint arXiv:2102.04306,

  5. [5]

    Algorithmic fairness in artificial in- telligence for medicine and healthcare

    Richard J Chen, Judy J Wang, Drew FK Williamson, Tiffany Y Chen, Jana Lipkova, Ming Y Lu, Sharifa Sahai, and Faisal Mahmood. Algorithmic fairness in artificial in- telligence for medicine and healthcare. Nature biomedical engineering, 7(6):719–742, 2023. 1

  6. [6]

    Measures of the amount of ecologic association between species

    Lee R Dice. Measures of the amount of ecologic association between species. Ecology, 26(3):297–302, 1945. 5, 3

  7. [7]

    FairTune: Optimizing Parameter Efficient Fine Tuning for Fairness in Medical Image Analysis

    Raman Dutt, Ondrej Bohdal, Sotirios A Tsaftaris, and Tim- othy Hospedales. Fairtune: Optimizing parameter efficient fine tuning for fairness in medical image analysis. arXiv preprint arXiv:2310.05055, 2023. 1

  8. [8]

    Sharpness-aware minimization for efficiently improving generalization

    Pierre Foret, Ariel Kleiner, Hossein Mobahi, and Behnam Neyshabur. Sharpness-aware minimization for efficiently improving generalization. arXiv preprint arXiv:2010.01412,

Show all 68 references
  1. [9]

    The use of ranks to avoid the assumption of normality implicit in the analysis of variance

    Milton Friedman. The use of ranks to avoid the assumption of normality implicit in the analysis of variance. Journal of the american statistical association, 32(200):675–701, 1937. 5

  2. [10]

    Physiobank, physiotoolkit, and physionet: compo- nents of a new research resource for complex physiologic signals

    Ary L Goldberger, Luis AN Amaral, Leon Glass, Jeffrey M Hausdorff, Plamen Ch Ivanov, Roger G Mark, Joseph E Mi- etus, George B Moody, Chung-Kang Peng, and H Eugene Stanley. Physiobank, physiotoolkit, and physionet: compo- nents of a new research resource for complex physiologi...

  3. [11]

    Knowledge distillation: A survey

    Jianping Gou, Baosheng Yu, Stephen J Maybank, and Dacheng Tao. Knowledge distillation: A survey. Interna- tional Journal of Computer Vision, 129(6):1789–1819, 2021. 3

  4. [12]

    Evaluating deep neural networks trained on clinical images in dermatology with the fitzpatrick 17k dataset

    Matthew Groh, Caleb Harris, Luis Soenksen, Felix Lau, Rachel Han, Aerin Kim, Arash Koochek, and Omar Badri. Evaluating deep neural networks trained on clinical images in dermatology with the fitzpatrick 17k dataset. In Proceed- ings of the IEEE/CVF Conference on Computer Visio...

  5. [13]

    FFB: A fair fairness benchmark for in-processing group fairness methods

    Xiaotian Han, Jianfeng Chi, Yu Chen, Qifan Wang, Han Zhao, Na Zou, and Xia Hu. FFB: A fair fairness benchmark for in-processing group fairness methods. In The Twelfth In- ternational Conference on Learning Representations , 2024. 1

  6. [14]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 4

  7. [15]

    Distilling the knowledge in a neural net- work

    Geoffrey Hinton. Distilling the knowledge in a neural net- work. arXiv preprint arXiv:1503.02531, 2015. 3, 4

  8. [16]

    Simple data balancing achieves com- petitive worst-group-accuracy

    Badr Youbi Idrissi, Martin Arjovsky, Mohammad Pezeshki, and David Lopez-Paz. Simple data balancing achieves com- petitive worst-group-accuracy. In Conference on Causal Learning and Reasoning, pages 336–351. PMLR, 2022. 1

  9. [17]

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

    Jeremy Irvin, Pranav Rajpurkar, Michael Ko, Yifan Yu, Sil- viana Ciurea-Ilcus, Chris Chute, Henrik Marklund, Behzad Haghgoo, Robyn Ball, Katie Shpanskaya, et al. Chexpert: A large chest radiograph dataset with uncertainty labels and expert comparison. In Proceedings of the AAA...

  10. [18]

    ´Etude comparative de la distribution florale dans une portion des alpes et des jura

    Paul Jaccard. ´Etude comparative de la distribution florale dans une portion des alpes et des jura. Bull Soc Vaudoise Sci Nat, 37:547–579, 1901. 5, 3

  11. [19]

    Mimic-cxr-jpg, a large publicly available database of labeled chest radiographs

    Alistair EW Johnson, Tom J Pollard, Nathaniel R Green- baum, Matthew P Lungren, Chih-ying Deng, Yifan Peng, Zhiyong Lu, Roger G Mark, Seth J Berkowitz, and Steven Horng. Mimic-cxr-jpg, a large publicly available database of labeled chest radiographs. arXiv preprint arXiv:1901.07042,

  12. [20]

    Achieving fairness in medical devices

    Achuta Kadambi. Achieving fairness in medical devices. Science, 372(6537):30–31, 2021. 1

  13. [21]

    Learning not to learn: Training deep neural networks with biased data

    Byungju Kim, Hyunwoo Kim, Kyungsu Kim, Sungjin Kim, and Junmo Kim. Learning not to learn: Training deep neural networks with biased data. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 9012–9020, 2019. 2

  14. [22]

    Multi- accuracy: Black-box post-processing for fairness in classifi- cation

    Michael P Kim, Amirata Ghorbani, and James Zou. Multi- accuracy: Black-box post-processing for fairness in classifi- cation. In Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society, pages 247–254, 2019. 2

  15. [23]

    Papila: Dataset with fundus images and clinical data of both eyes of the same patient for glaucoma assessment

    Oleksandr Kovalyk, Juan Morales-S ´anchez, Rafael Verd ´u- Monedero, Inmaculada Sell ´es-Navarro, Ana Palaz ´on- Cabanes, and Jos ´e-Luis Sancho-G ´omez. Papila: Dataset with fundus images and clinical data of both eyes of the same patient for glaucoma assessment. Scientific D...

  16. [24]

    Gender imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis

    Agostina J Larrazabal, Nicol ´as Nieto, Victoria Peterson, Diego H Milone, and Enzo Ferrante. Gender imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis. Proceedings of the National Academy of Sciences, 117(23):12592–12594, 2020. 2

  17. [25]

    Asymmetric temper- ature scaling makes larger networks teach well again

    Xin-Chun Li, Wen-Shu Fan, Shaoming Song, Yinchuan Li, Shao Yunfeng, De-Chuan Zhan, et al. Asymmetric temper- ature scaling makes larger networks teach well again. Ad- vances in neural information processing systems , 35:3830– 3842, 2022. 4

  18. [26]

    Bias mitigation post-processing for individual and group fairness

    Pranay K Lohia, Karthikeyan Natesan Ramamurthy, Man- ish Bhide, Diptikalyan Saha, Kush R Varshney, and Ruchir Puri. Bias mitigation post-processing for individual and group fairness. In Icassp 2019-2019 ieee international con- ference on acoustics, speech and signal processing...

  19. [27]

    Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks

    Jiasen Lu, Dhruv Batra, Devi Parikh, and Stefan Lee. Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks. Advances in neural information processing systems, 32, 2019. 3

  20. [28]

    Harvard glaucoma detection and progression: A multimodal multitask dataset and generalization-reinforced semi-supervised learning

    Yan Luo, Min Shi, Yu Tian, Tobias Elze, and Mengyu Wang. Harvard glaucoma detection and progression: A multimodal multitask dataset and generalization-reinforced semi-supervised learning. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 20471– ...

  21. [29]

    Eye fairness: A large-scale 3d imaging dataset for equi- table eye diseases screening and fair identity scaling

    Yan Luo, Yu Tian, Min Shi, Tobias Elze, and Mengyu Wang. Eye fairness: A large-scale 3d imaging dataset for equi- table eye diseases screening and fair identity scaling. arXiv preprint arXiv:2310.02492, 2023

  22. [30]

    Harvard glaucoma fairness: A retinal nerve disease dataset for fairness learning and fair identity normalization

    Yan Luo, Yu Tian, Min Shi, Tobias Elze, and Mengyu Wang. Harvard glaucoma fairness: A retinal nerve disease dataset for fairness learning and fair identity normalization. arXiv preprint arXiv:2306.09264, 2023. 2

  23. [31]

    Fairvi- sion: Equitable deep learning for eye disease screening via fair identity scaling, 2024

    Yan Luo, Muhammad Osama Khan, Yu Tian, Min Shi, Ze- hao Dou, Tobias Elze, Yi Fang, and Mengyu Wang. Fairvi- sion: Equitable deep learning for eye disease screening via fair identity scaling, 2024. 1, 2, 3, 4, 5, 6, 7

  24. [32]

    Fairclip: Har- nessing fairness in vision-language learning

    Yan Luo, Min Shi, Muhammad Osama Khan, Muham- mad Muneeb Afzal, Hao Huang, Shuaihang Yuan, Yu Tian, Luo Song, Ava Kouhana, Tobias Elze, et al. Fairclip: Har- nessing fairness in vision-language learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- ter...

  25. [33]

    Learning adversarially fair and transferable represen- tations

    David Madras, Elliot Creager, Toniann Pitassi, and Richard Zemel. Learning adversarially fair and transferable represen- tations. In International Conference on Machine Learning , pages 3384–3393. PMLR, 2018. 1, 2, 6, 7

  26. [34]

    Systematic outperformance of 112 der- matologists in multiclass skin cancer image classification by convolutional neural networks.European Journal of Cancer, 119:57–65, 2019

    Roman C Maron, Michael Weichenthal, Jochen S Utikal, Achim Hekler, Carola Berking, Axel Hauschild, Alexan- der H Enk, Sebastian Haferkamp, Joachim Klode, Dirk Schadendorf, et al. Systematic outperformance of 112 der- matologists in multiclass skin cancer image classification b...

  27. [35]

    Minimax pareto fairness: A multi objective perspective

    Natalia Martinez, Martin Bertran, and Guillermo Sapiro. Minimax pareto fairness: A multi objective perspective. In International conference on machine learning, pages 6755–

  28. [36]

    Whinston, Jerry R

    Andreu Mas-Colell, Michael D. Whinston, Jerry R. Green, et al. Microeconomic theory, volume 1 . Oxford university press New York, 1995. 2

  29. [37]

    A survey on bias and fairness in machine learning

    Ninareh Mehrabi, Fred Morstatter, Nripsuta Saxena, Kristina Lerman, and Aram Galstyan. A survey on bias and fairness in machine learning. ACM computing surveys (CSUR) , 54 (6):1–35, 2021. 1

  30. [38]

    Distribution-free multiple compar- isons

    Peter Bjorn Nemenyi. Distribution-free multiple compar- isons. Princeton University, 1963. 5

  31. [39]

    Smote: synthetic minority over-sampling technique

    V Chawla Nitesh. Smote: synthetic minority over-sampling technique. J Artif Intell Res, 16(1):321, 2002. 1, 2

  32. [40]

    Dissecting racial bias in an algorithm used to manage the health of populations

    Ziad Obermeyer, Brian Powers, Christine V ogeli, and Send- hil Mullainathan. Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464): 447–453, 2019. 1, 2

  33. [41]

    Fair contrastive learn- ing for facial attribute classification

    Sungho Park, Jewook Lee, Pilhyeon Lee, Sunhee Hwang, Dohyung Kim, and Hyeran Byun. Fair contrastive learn- ing for facial attribute classification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10389–10398, 2022. 2

  34. [42]

    On fairness and calibration

    Geoff Pleiss, Manish Raghavan, Felix Wu, Jon Kleinberg, and Kilian Q Weinberger. On fairness and calibration. Ad- vances in neural information processing systems , 30, 2017. 2

  35. [43]

    Discovering fair representations in the data domain

    Novi Quadrianto, Viktoriia Sharmanska, and Oliver Thomas. Discovering fair representations in the data domain. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8227–8236, 2019. 2

  36. [44]

    Fair attribute classification through latent space de-biasing

    Vikram V Ramaswamy, Sunnie SY Kim, and Olga Rus- sakovsky. Fair attribute classification through latent space de-biasing. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9301–9310, 2021

  37. [45]

    Fr-train: A mutual information-based approach to fair and robust training

    Yuji Roh, Kangwook Lee, Steven Whang, and Changho Suh. Fr-train: A mutual information-based approach to fair and robust training. In International Conference on Machine Learning, pages 8147–8157. PMLR, 2020. 2

  38. [46]

    Distributionally robust neural networks for group shifts: On the importance of regularization for worst- case generalization

    Shiori Sagawa, Pang Wei Koh, Tatsunori B Hashimoto, and Percy Liang. Distributionally robust neural networks for group shifts: On the importance of regularization for worst- case generalization. arXiv preprint arXiv:1911.08731, 2019. 1, 2, 6, 7

  39. [47]

    Distilbert, a distilled version of bert: Smaller, faster, cheaper and lighter

    V Sanh. Distilbert, a distilled version of bert: Smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108, 2019. 3

  40. [48]

    Fairness by learning orthogonal disentan- gled representations

    Mhd Hasan Sarhan, Nassir Navab, Abouzar Eslami, and Shadi Albarqouni. Fairness by learning orthogonal disentan- gled representations. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXIX 16, pages 746–761. Springer, 2020. 1, 2

  41. [49]

    Equitable ar- tificial intelligence for glaucoma screening with fair identity normalization

    Min Shi, Yan Luo, Yu Tian, Lucy Q Shen, Tobias Elze, Nazlee Zebardast, Mohammad Eslami, Saber Kazeminasab, Michael V Boland, David S Friedman, et al. Equitable ar- tificial intelligence for glaucoma screening with fair identity normalization. medRxiv, pages 2023–12, 2023

  42. [50]

    Equitable deep learning for diabetic retinopathy detection using multi-dimensional reti- nal imaging with fair adaptive scaling: a retrospective study

    Min Shi, Muhammad Muneeb Afzal, Hao Huang, Congcong Wen, Yan Luo, Muhammad Osama Khan, Yu Tian, Leo Kim, Tobias Elze, Yi Fang, et al. Equitable deep learning for diabetic retinopathy detection using multi-dimensional reti- nal imaging with fair adaptive scaling: a retrospectiv...

  43. [51]

    A method of establishing groups of equal amplitude in plant sociology based on similarity of species content and its application to analyses of the vegeta- tion on danish commons

    Thorvald Sorensen. A method of establishing groups of equal amplitude in plant sociology based on similarity of species content and its application to analyses of the vegeta- tion on danish commons. Biologiske skrifter, 5:1–34, 1948. 5, 3

  44. [52]

    End: Entangling and disentangling deep rep- resentations for bias correction

    Enzo Tartaglione, Carlo Alberto Barbano, and Marco Grangetto. End: Entangling and disentangling deep rep- resentations for bias correction. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13508–13517, 2021. 1, 2

  45. [53]

    Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results

    Antti Tarvainen and Harri Valpola. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. Advances in neural information processing systems, 30, 2017. 3

  46. [54]

    Fairseg: A large-scale medical image seg- mentation dataset for fairness learning using segment any- thing model with fair error-bound scaling

    Yu Tian, Min Shi, Yan Luo, Ava Kouhana, Tobias Elze, and Mengyu Wang. Fairseg: A large-scale medical image seg- mentation dataset for fairness learning using segment any- thing model with fair error-bound scaling. In The Twelfth In- ternational Conference on Learning Represent...

  47. [55]

    Fairdomain: Achieving fairness in cross-domain medical image segmentation and classifica- tion

    Yu Tian, Congcong Wen, Min Shi, Muhammad Muneeb Afzal, Hao Huang, Muhammad Osama Khan, Yan Luo, Yi Fang, and Mengyu Wang. Fairdomain: Achieving fairness in cross-domain medical image segmentation and classifica- tion. arXiv preprint arXiv:2407.08813, 2024. 2

  48. [56]

    The ham10000 dataset, a large collection of multi-source der- matoscopic images of common pigmented skin lesions

    Philipp Tschandl, Cliff Rosendahl, and Harald Kittler. The ham10000 dataset, a large collection of multi-source der- matoscopic images of common pigmented skin lesions. Sci- entific data, 5(1):1–9, 2018. 1, 2, 5

  49. [57]

    An overview of statistical learning the- ory

    Vladimir N Vapnik. An overview of statistical learning the- ory. IEEE transactions on neural networks, 10(5):988–999,

  50. [58]

    Fairness-aware ad- versarial perturbation towards bias mitigation for deployed deep models

    Zhibo Wang, Xiaowei Dong, Henry Xue, Zhifei Zhang, Weifeng Chiu, Tao Wei, and Kui Ren. Fairness-aware ad- versarial perturbation towards bias mitigation for deployed deep models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10379–...

  51. [59]

    Cutmix: Regu- larization strategy to train strong classifiers with localizable features

    Sangdoo Yun, Dongyoon Han, Seong Joon Oh, Sanghyuk Chun, Junsuk Choe, and Youngjoon Yoo. Cutmix: Regu- larization strategy to train strong classifiers with localizable features. In Proceedings of the IEEE/CVF international con- ference on computer vision, pages 6023–6032, 2019. 5, 7

  52. [60]

    Fairness constraints: Mechanisms for fair classification

    Muhammad Bilal Zafar, Isabel Valera, Manuel Gomez Ro- griguez, and Krishna P Gummadi. Fairness constraints: Mechanisms for fair classification. In Artificial intelligence and statistics, pages 962–970. PMLR, 2017. 2

  53. [61]

    Mitigating unwanted biases with adversarial learning

    Brian Hu Zhang, Blake Lemoine, and Margaret Mitchell. Mitigating unwanted biases with adversarial learning. In Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society, pages 335–340, 2018. 2

  54. [62]

    Towards accuracy-fairness para- dox: Adversarial example-based data augmentation for vi- sual debiasing

    Yi Zhang and Jitao Sang. Towards accuracy-fairness para- dox: Adversarial example-based data augmentation for vi- sual debiasing. In Proceedings of the 28th ACM Interna- tional Conference on Multimedia , pages 4346–4354, 2020. 2

  55. [63]

    Conditional learning of fair representations

    Han Zhao, Amanda Coston, Tameem Adel, and Geoffrey J Gordon. Conditional learning of fair representations. arXiv preprint arXiv:1910.07162, 2019. 1, 2

  56. [64]

    Medfair: Benchmarking fairness for medical imaging

    Yongshuo Zong, Yongxin Yang, and Timothy Hospedales. Medfair: Benchmarking fairness for medical imaging. arXiv preprint arXiv:2210.01725, 2022. 1, 2, 5, 7 Fair Distillation: Teaching Fairness from Biased Teachers in Medical Imaging Supplementary Material

  57. [65]

    For data splitting

    Data Pre-processing The data preprocessing in this work follows the approach outlined in [64]. For data splitting. The dataset is randomly partitioned into training, validation, and testing sets, main- taining an 80/10/10 ratio unless stated otherwise. We then binarize both th...

  58. [66]

    All datasets are publicly available and can be accessed through the URLs listed in Table 4

    Dataset Details In this section, we provide details about the datasets. All datasets are publicly available and can be accessed through the URLs listed in Table 4. The dataset statistics are pro- vided in Table 5. Additionally, we summarize the statistics for the subgroups and...

  59. [67]

    This measure promotes fairness by proportionally reducing the overall AUC in response to increased subgroup disparities

    Fairness Measures For the classification metrics, the ES-AUC provides an equity-focused assessment by scaling the overall AUC based on performance differences across subgroups, calcu- lated as follows: ES − AUC = AUC(D) 1 + 1 A P g∈A |AUC(D) − AUC(Dg)| (4) where AUC(D) represe...

  60. [68]

    11 shows the complete set of classification results, while Tabs

    Complete Set of Results Tab. 11 shows the complete set of classification results, while Tabs. 12 to 14 show a complete set of segmentation results. Table 11. Complete evaluation of fairness classification models across medical imaging benchmarks: HAM10000, Fitzpatrick17k, PA- ...

Pith tools

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