Pith. sign in

REVIEW 4 major objections 5 minor 60 references

Constructing Fair Latent Space for Intersection of Fairness and Explainability

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read The paper argues that training a small invertible neural network on the frozen latent space of a pretrained generative model creates a fair latent space in which labels and sensitive attributes are disentangled, and that this space both…

desk verdict Useful empirical plug-in for fair generative latent spaces with a theory section that needs serious patching before the paper is citable for its claims. read the letter →

arxiv 2412.17523 v2 pith:KK65YXUN submitted 2024-12-23 cs.LG cs.AIcs.CV

classification cs.LGcs.AIcs.CV
keywords fairlatentspacecounterfactualexplanationsinvertibleneuralnetworksgenerativemodelfairnessgroupmetricsequalizedoddsdemographicparitydisentangledrepresentationlearning
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 tries to establish that a small invertible neural network, inserted between the encoder and decoder of a pretrained generative model, can reorganize the model's latent space so that the information used for the task label and the information about a sensitive attribute live in separate coordinate blocks. The payoff is that a frozen generator becomes both fairer and explainable: fairness metrics improve without retraining the generator, and counterfactual images can be produced by moving along the label or sensitive-attribute directions. On CelebA attractiveness classification with gender as the sensitive attribute, the method reports an 86.8% reduction in equalized odds, a 52.0% reduction in demographic parity, and a 39.9% increase in worst-group accuracy. The paper also presents counterfactual generations showing that the attractiveness direction no longer correlates with gender, and that training the module is far cheaper than retraining the generator.

What carries the argument

The load-bearing object is the invertible neural network $f_{\theta}$ mapping the frozen encoder output $E$ to the fair code $Z = [Z^Y, Z^S]$; invertibility lets changes in the fair space be decoded back into images by the frozen generator. The argument is carried by two theoretical results. Theorem 1 recasts an information-bottleneck objective as $E[\log \det C_{Z^Y|Y}] - \lambda \log \det C_{Z^Y}$, whose minimization is achieved by making the covariance matrix of $Z^Y$ a scalar matrix, which the losses $L_{dg}$ and $L_{eq}$ enforce directly. Theorem 2 uses a noise-contrastive estimate of mutual information between opposite sensitive-attribute groups to claim that, once the covariance is scalar, maximizing that mutual information reduces to minimizing the $L_2$ distance between the two groups; the paper implements this as the mask-based distance loss $L_{di}$. The classifier weight vectors trained on $Z^Y$ and $Z^S$ serve as the axes for counterfactual manipulation.

What would settle it

On a trained model, collect the norms of the label-block representations across a batch and check whether they are close to a common constant; if the spread is large, the derivation of Theorem 2 does not apply, and one should test whether the distance loss still improves fairness when each sample's norm is explicitly normalized before computing the loss.

Watch

Extended reading notes

Core claim

The central claim is that fairness and explainability can be obtained jointly by constructing a fair latent space rather than by changing the predictor. The paper defines a fair latent space as one in which the representation decomposes as $Z = [Z^Y, Z^S]$, where the label block $Z^Y$ carries only label information and the sensitive-attribute block $Z^S$ carries only sensitive-attribute information, and then reports that training an invertible network to enforce this structure makes the latent space fair by standard group metrics. The mechanism is a combination of losses: a diagonalizing loss and an equalizing loss turn the covariance matrix of each block into a scalar multiple of the identity, a distance loss pulls same-label samples with different sensitive attributes together while pushing same-sensitive samples with different labels apart, and a Gaussianizing log-likelihood loss keeps the invertible mapping well behaved. With this space in place, counterfactual explanations are generated by shifting a latent code along the classifier weight vector for the label or for the sensitive attribute, and the paper claims the resulting counterfactual images visibly preserve the manipulated attribute while removing gender dependence.

Load-bearing premise

The theoretical link between the distance loss and mutual information maximization assumes that after the covariance matrix is made scalar, every individual image's latent vector has almost the same length; a scalar covariance matrix only forces the average per-dimension variance across the batch to be equal, not each sample's norm.

Editorial extensions

If this is right

  • If the central claim holds, any pretrained encoder-decoder generative model can be retrofitted for fairness by training only a small invertible module, avoiding the cost of retraining the generator.
  • The resulting latent space supports two kinds of counterfactual explanations: label counterfactuals that show what drove a decision, and sensitive-attribute counterfactuals that show whether the decision would survive a change in the sensitive attribute.
  • The reported numbers imply that group-fairness metrics such as equalized odds and demographic parity can be improved without sacrificing worst-group accuracy, since worst-group accuracy rises rather than falls in the reported settings.
  • Because the module operates on the latent space rather than on pixels, its training cost scales with latent dimensionality and dataset size, not image resolution, giving a large reported speedup on high-resolution data.

Reading between the lines

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

  • A testable extension is to apply the same plug-in module to other sensitive attributes, such as race or age, and check whether the measured fairness gains transfer, since the paper's multi-attribute experiment is the only such check.
  • If the fair-space directions are truly semantic, the same axis manipulation could double as a general controllable image-editing tool; the paper does not claim this, but the mechanism implies it.
  • The proof of Theorem 2 would be stronger if the batch were normalized so every sample has the same latent norm; without that normalization, the empirical gains from the distance loss may come from the loss itself rather than from the mutual-information justification, which is directly testable by comparing normalized and unnormalized variants.
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

4 major / 5 minor

Summary. The paper proposes a plug-in module for pretrained generative models that maps the frozen latent space to a 'fair latent space' using an invertible neural network (Glow) trained on the latent codes of a Diffusion Autoencoder. The training objective combines a covariance-diagonalization loss, a variance-equalization loss, a distance-based contrastive loss, a Gaussianizing normalizing-flow loss, and a classifier loss. The authors claim that this disentangles label and sensitive-attribute information, improves group-fairness metrics (EO, DP, WGA) across CelebA, CelebAHQ, and UTK Face, and enables counterfactual explanations by shifting latent codes along classifier weight directions. The theoretical motivation is an information-bottleneck argument (Theorem 1) and a noise-contrastive-estimation argument (Theorem 2) intended to justify the three fair-loss components.

Significance. If the claims hold, the contribution is practically valuable: a fairness module that attaches to a frozen generative model, requires training only a small INN, and provides counterfactual explanations, with large reported reductions in EO/DP and gains in WGA across multiple datasets and sensitive-attribute settings. The paper is strong on the empirical side: it reports ablations for each loss component (Tables 3 and 5-6), compares against several representation-learning baselines, includes a quantitative counterfactual evaluation using CLIP, and documents computational savings (Table 7). The main weakness is the theoretical narrative: the proof of Theorem 2 contains a genuine gap, and there are additional unproven assumptions in the derivation of Theorem 1 and the Gaussianizing loss. These issues do not necessarily invalidate the empirical method, but they do undermine the paper's central claim that the losses are derived from the theory, so the manuscript needs substantial revision.

major comments (4)
  1. [Appendix A.2, Eq. (18)] The proof of Theorem 2 hinges on the approximation ||z_i||^2 ≈ d_y c = R. This does not follow from the preceding covariance argument: a scalar covariance C(Z^Y) = cI only fixes the batch-averaged squared norm (1/n)Σ_i ||z_i||^2 = d_y c, not each sample's norm. The substitution in Eq. (19), which converts the positive-pair inner product into R − ½L2², is therefore unsupported. In fact, the Gaussianizing loss Lg (Eq. (9)) drives z toward a standard normal, so per-sample norms follow a chi distribution with variance proportional to d_y, not a constant. Because Ldi is presented as the practical realization of Theorem 2, the theoretical link between Ldi and mutual information maximization is not established. The authors should either prove a concentration bound for ||z_i||² under the training losses or explicitly reframe Ldi as an empirically motivated heuristic.
  2. [Section 3.1, Theorem 1] The text states that the term E_Y[log det(C_{Z^Y|Y})] remains constant when optimizing Z^Y, and therefore only the second term log det(C_ZY) needs to be minimized. This constancy is not proven and is generally false: the losses Ldg and Leq in Eqs. (5)–(6) change the conditional covariance structure of Z^Y given Y. Without an additional argument bounding or fixing this term, the reduction of the information-bottleneck objective to covariance diagonalization and equalization is incomplete. The empirical ablation in Table 3 supports the usefulness of Ldg and Leq, but the theoretical justification as written does not.
  3. [Section 3.3, Eq. (9)] The stated loss Lg = −(1/n)Σ(||fθ(e)||² + log|det J_fθ(e)|) is not a negative log-likelihood for a standard Gaussian prior; the standard NLL is (1/2)||fθ(e)||² − log|det J_fθ(e)| + const. Minimizing the expression as written would push latent norms to grow, contrary to the Gaussianization claim and to the assumption used in Theorem 1 and in Eq. (18). Please correct the sign (and factor) or clarify the convention; this is material because the Gaussian assumption underpins the theoretical derivations.
  4. [Appendix A.2, Eq. (20)] Even if the constant-norm approximation were granted, the bound on the negative-pair term is incomplete: the derivation leaves the term (z_{s0,i})^T z_{s0,j} unaccounted for, so the denominator of the NCE bound in Eq. (16) is not shown to be controlled by cross-group L2 distances alone. The conclusion that minimizing L2 distance between groups maximizes the mutual information lower bound therefore still lacks a complete proof.
minor comments (5)
  1. [Appendix B heading] The header 'Experiemental details' should be corrected to 'Experimental details'.
  2. [Section 3.2, Eq. (6)] The target variance c in the equalizing loss Leq is not reported in the hyperparameter list in Appendix B.5; without this value the loss is not fully specified for reproduction.
  3. [Section 4.1, Eq. (11)] The symbol y_p in the demographic parity formula is used without an explicit definition; please define it as the positive class.
  4. [Tables 2 and 6] The FSCL row has no entries for the multi-sensitive CelebA setting; please state whether FSCL was not run or did not converge in that setting.
  5. [Section 4.4] The choice to move latent codes along the normalized classifier weight vector h/||h|| is presented as 'the best choice' without justification; if this is a design heuristic, it should be labeled as such in the text.

Circularity Check

0 steps flagged · score 2.0 of 10

No material circularity: the fairness results are measured on held-out test data against external baselines, and the theoretical losses are not fitted to the reported metrics.

full rationale

The paper's headline fairness claims (EO, DP, WGA reductions) are empirical test-set evaluations compared with external baselines (SimCLR, SupCon, FSCL) and with ablated variants of the proposed module; no reported fairness number is a fitted constant that is then relabeled as a prediction. The loss functions (Ldg, Leq, Ldi, Lg, Lcls) are derived from stated assumptions including Gaussian representations, scalar covariance, and the NCE lower bound, and the ablation studies test the contribution of each component rather than tuning the components to the target metrics. The self-citations (Joo et al. 2023; Han et al. 2024) appear only as motivational or related-work references and are not load-bearing for the central derivation: no premise of the method relies on an unverified result from the authors' own prior work, and no uniqueness theorem is imported from those papers. The proof gap in Theorem 2 (Appendix A.2, Eq. 18), where a scalar covariance matrix is used to assert the per-sample norm approximation ||z_i||^2 = d_y c = R, is a mathematical correctness concern about an approximation, not circularity: the theorem's conclusion is not assumed among its premises, and the practical loss is evaluated empirically. Overall, the derivation chain is not equivalent to its inputs by construction, so no significant circularity is present.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central empirical claim rests on several modeling assumptions and hand-chosen hyperparameters. The most fragile is the per-sample norm approximation in the proof of Theorem 2, which is not implied by the scalar covariance objective. The Gaussian and constancy assumptions in Theorem 1 are partially self-imposed through the Gaussianizing loss.

free parameters (6)
  • λdg (covariance diagonalization weight) = 1
    Chosen by hand in Appendix B.5; the fairness results depend on this weight.
  • λeq (variance equalization weight) = 10
    Chosen by hand; controls how strongly diagonal variances are equalized.
  • λdi (distance contrastive weight) = 1 or 3
    Set to 1 or 3 depending on setting; directly controls the Ldi loss that implements Theorem 2.
  • λcls (classifier loss weight) = 1 or 3
    Weight for cross-entropy on Z^Y and Z^S.
  • c (target variance in Leq) = not stated (likely 1)
    Variance target for each dimension; the paper does not give the value used.
  • Counterfactual shift scale = 3
    Appendix D: counterfactuals are generated by tripling the unit classifier weight vector; the choice is arbitrary and not justified.
assumptions (5)
  • domain assumption The representation Z^Y follows a Gaussian distribution (Theorem 1).
    Used to express entropy as log determinant of covariance in Eq. (14); the training loss Lg in Eq. (9) is designed to enforce a standard Gaussian prior, so the assumption is partially self-imposed.
  • ad hoc to paper The first term E_Y[log det(C_{Z^Y|Y})] remains constant when optimizing Z^Y (Sec. 3.1).
    Needed to reduce the IB loss to maximizing log det(C_{Z^Y}); no proof is provided for constancy.
  • ad hoc to paper Per-sample latent norm is approximately constant, ||z_i||^2 ≈ R (Appendix A.2, Eq. 18).
    Used to rewrite the NCE inner product as a negative squared L2 distance; not implied by scalar covariance.
  • domain assumption The pre-trained DiffAE latent space is semantically meaningful and preserves information needed for fair classification after invertible transformation.
    The module only trains an INN on frozen embeddings; if the embeddings lack label/sensitive information or are not smoothly manipulable, the method fails.
  • ad hoc to paper Moving a latent code along the classifier weight vector h/||h|| produces meaningful counterfactual explanations (Sec. 4.4).
    No proof or validation that this direction corresponds to the model's actual decision boundary; the scale factor is chosen as 3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Constructing Fair Latent Space for Intersection of Fairness and Explainability." pith.science (2026). https://pith.science/paper/KK65YXUN

@misc{pith2026241217523,
  author       = {Pith},
  title        = {Pith review of: Constructing Fair Latent Space for Intersection of Fairness and Explainability},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KK65YXUN}},
  note         = {Machine review of arXiv:2412.17523}
}
read the original abstract

As the use of machine learning models has increased, numerous studies have aimed to enhance fairness. However, research on the intersection of fairness and explainability remains insufficient, leading to potential issues in gaining the trust of actual users. Here, we propose a novel module that constructs a fair latent space, enabling faithful explanation while ensuring fairness. The fair latent space is constructed by disentangling and redistributing labels and sensitive attributes, allowing the generation of counterfactual explanations for each type of information. Our module is attached to a pretrained generative model, transforming its biased latent space into a fair latent space. Additionally, since only the module needs to be trained, there are advantages in terms of time and cost savings, without the need to train the entire generative model. We validate the fair latent space with various fairness metrics and demonstrate that our approach can effectively provide explanations for biased decisions and assurances of fairness.

Figures

Figures reproduced from arXiv: 2412.17523 by the authors.

Figure 1
Figure 1. (A) Models aimed at enhancing fairness without any explanation. (B) The proposed model trains an invertible neural [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of our approach connecting theoretical analysis to practical implementation, comprising three main com [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Counterfactual explanations with samples initially misclassified as unattractive by the original model. The x-axis [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: (Left) Gender misclassification rates when rep [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Counterfactual explanations for samples correctly classified by our model with the label [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 42 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

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

  3. [3]

    A.; and Gokhale, D

    Ahmed, N. A.; and Gokhale, D. 1989. Entropy expressions and their estimators for multivariate distributions. IEEE Transactions on Information Theory, 35(3): 688--692

  4. [4]

    Bardes, A.; Ponce, J.; and LeCun, Y. 2022. VICR eg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning. In International Conference on Learning Representations

  5. [5]

    Barocas, S.; and Selbst, A. D. 2016. Big data's disparate impact. Calif. L. Rev., 104: 671

  6. [6]

    Begley, T.; Schwedes, T.; Frye, C.; and Feige, I. 2020. Explainability for fair machine learning. arXiv preprint arXiv:2010.07389

  7. [7]

    Behrmann, J.; Vicol, P.; Wang, K.-C.; Grosse, R.; and Jacobsen, J.-H. 2021. Understanding and mitigating exploding inverses in invertible neural networks. In International Conference on Artificial Intelligence and Statistics, 1792--1800. PMLR

  8. [8]

    Brown, C. 2003. Giving up levelling down. Economics & Philosophy, 19(1): 111--134

Show all 60 references
  1. [9]

    Caton, S.; and Haas, C. 2024. Fairness in machine learning: A survey. ACM Computing Surveys, 56(7): 1--38

  2. [10]

    Chen, T.; Kornblith, S.; Norouzi, M.; and Hinton, G. 2020. A simple framework for contrastive learning of visual representations. In International conference on machine learning, 1597--1607. PMLR

  3. [11]

    Chiappa, S. 2019. Path-specific counterfactual fairness. In Proceedings of the AAAI conference on artificial intelligence, volume 33, 7801--7808

  4. [12]

    Cho, J.; Zala, A.; and Bansal, M. 2023. Dall-eval: Probing the reasoning skills and social biases of text-to-image generation models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 3043--3054

  5. [13]

    Chouldechova, A. 2017. Fair prediction with disparate impact: A study of bias in recidivism prediction instruments. Big data, 5(2): 153--163

  6. [14]

    Christiano, T.; and Braynen, W. 2008. Inequality, injustice and levelling down. Ratio, 21(4): 392--420

  7. [15]

    Creager, E.; Madras, D.; Jacobsen, J.-H.; Weis, M.; Swersky, K.; Pitassi, T.; and Zemel, R. 2019. Flexibly fair representation learning by disentanglement. In International conference on machine learning, 1436--1445. PMLR

  8. [16]

    N.; and Sharma, A

    Dash, S.; Balasubramanian, V. N.; and Sharma, A. 2022. Evaluating and mitigating bias in image classifiers: A causal perspective using counterfactuals. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 915--924

  9. [17]

    Denton, E.; Hutchinson, B.; Mitchell, M.; Gebru, T.; and Zaldivar, A. 2019. Image counterfactual sensitivity analysis for detecting unintended bias. arXiv preprint arXiv:1906.06439

  10. [18]

    S.; and Pontil, M

    Donini, M.; Oneto, L.; Ben-David, S.; Shawe-Taylor, J. S.; and Pontil, M. 2018. Empirical risk minimization under fairness constraints. Advances in neural information processing systems, 31

  11. [19]

    Doran, B. 2001. Reconsidering the levelling-down objection against egalitarianism. Utilitas, 13(1): 65--85

  12. [20]

    Esser, P.; Rombach, R.; and Ommer, B. 2020. A disentangling invertible interpretation network for explaining latent representations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 9223--9232

  13. [21]

    Gustafson, L.; Rolland, C.; Ravi, N.; Duval, Q.; Adcock, A.; Fu, C.-Y.; Hall, M.; and Ross, C. 2023. Facet: Fairness in computer vision evaluation benchmark. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 20370--20382

  14. [22]

    Han, H.; Kim, S.; Joo, H.; Hong, S.; and Lee, J. 2024. Mitigating Spurious Correlations via Disagreement Probability. arXiv preprint arXiv:2411.01757

  15. [23]

    Hardt, M.; Price, E.; and Srebro, N. 2016. Equality of opportunity in supervised learning. Advances in neural information processing systems, 29

  16. [24]

    Hong, S.; Park, I.; and Chun, S. Y. 2023. On the robustness of normalizing flows for inverse problems in imaging. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 10745--10755

  17. [25]

    Jacovi, A.; Marasovi \'c , A.; Miller, T.; and Goldberg, Y. 2021. Formalizing trust in artificial intelligence: Prerequisites, causes and goals of human trust in AI. In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency, 624--635

  18. [26]

    Joo, H.; Kim, J.; Han, H.; and Lee, J. 2023. Distributional Prototypical Methods for Reliable Explanation Space Construction. IEEE Access, 11: 34821--34834

  19. [27]

    Joo, J.; and K \"a rkk \"a inen, K. 2020. Gender slopes: Counterfactual fairness for computer vision models by attribute manipulation. In Proceedings of the 2nd international workshop on fairness, accountability, transparency and ethics in multimedia, 1--5

  20. [28]

    Karras, T.; Aila, T.; Laine, S.; and Lehtinen, J. 2018. Progressive Growing of GAN s for Improved Quality, Stability, and Variation. In International Conference on Learning Representations

  21. [29]

    Kehrenberg, T.; Bartlett, M.; Thomas, O.; and Quadrianto, N. 2020. Null-sampling for interpretable and fair representations. In Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XXVI 16, 565--580. Springer

  22. [30]

    Khosla, P.; Teterwak, P.; Wang, C.; Sarna, A.; Tian, Y.; Isola, P.; Maschinot, A.; Liu, C.; and Krishnan, D. 2020. Supervised contrastive learning. Advances in neural information processing systems, 33: 18661--18673

  23. [31]

    Kim, H.; and Mnih, A. 2018. Disentangling by factorising. In International conference on machine learning, 2649--2658. PMLR

  24. [32]

    P.; and Dhariwal, P

    Kingma, D. P.; and Dhariwal, P. 2018. Glow: Generative flow with invertible 1x1 convolutions. Advances in neural information processing systems, 31

  25. [33]

    Kirichenko, P.; Izmailov, P.; and Wilson, A. G. 2020. Why normalizing flows fail to detect out-of-distribution data. Advances in neural information processing systems, 33: 20578--20589

  26. [34]

    Kong, L.; de Masson d'Autume, C.; Yu, L.; Ling, W.; Dai, Z.; and Yogatama, D. 2020. A Mutual Information Maximization Perspective of Language Representation Learning. In International Conference on Learning Representations

  27. [35]

    Liu, J.; Li, Z.; Yao, Y.; Xu, F.; Ma, X.; Xu, M.; and Tong, H. 2022. Fair representation learning: An alternative to mutual information. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 1088--1097

  28. [36]

    Liu, Z.; Luo, P.; Wang, X.; and Tang, X. 2015. Deep learning face attributes in the wild. In Proceedings of the IEEE international conference on computer vision, 3730--3738

  29. [37]

    Oord, A. v. d.; Li, Y.; and Vinyals, O. 2018. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748

  30. [38]

    Park, S.; Lee, J.; Lee, P.; Hwang, S.; Kim, D.; and Byun, H. 2022. Fair contrastive learning for facial attribute classification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 10389--10398

  31. [39]

    Plecko, D.; and Bareinboim, E. 2023. Causal Fairness for Outcome Control. In Thirty-seventh Conference on Neural Information Processing Systems

  32. [40]

    Poole, B.; Ozair, S.; Van Den Oord, A.; Alemi, A.; and Tucker, G. 2019. On variational bounds of mutual information. In International Conference on Machine Learning, 5171--5180. PMLR

  33. [41]

    Preechakul, K.; Chatthee, N.; Wizadwongsa, S.; and Suwajanakorn, S. 2022. Diffusion autoencoders: Toward a meaningful and decodable representation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 10619--10629

  34. [42]

    Qiang, Y.; Li, C.; Brocanelli, M.; and Zhu, D. 2022. Counterfactual Interpolation Augmentation (CIA): A Unified Approach to Enhance Fairness and Explainability of DNN. In IJCAI, 732--739

  35. [43]

    V.; Kim, S

    Ramaswamy, V. V.; Kim, S. S.; and Russakovsky, O. 2021. Fair attribute classification through latent space de-biasing. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 9301--9310

  36. [44]

    M.; and Crawford, K

    Richardson, R.; Schultz, J. M.; and Crawford, K. 2019. Dirty data, bad predictions: How civil rights violations impact police data, predictive policing systems, and justice. NYUL Rev. Online, 94: 15

  37. [45]

    Rudin, C. 2019. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature machine intelligence, 1(5): 206--215

  38. [46]

    W.; Hashimoto, T

    Sagawa*, S.; Koh*, P. W.; Hashimoto, T. B.; and Liang, P. 2020. Distributionally Robust Neural Networks. In International Conference on Learning Representations

  39. [47]

    H.; Navab, N.; Eslami, A.; and Albarqouni, S

    Sarhan, M. H.; Navab, N.; Eslami, A.; and Albarqouni, S. 2020. Fairness by learning orthogonal disentangled representations. In Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XXIX 16, 746--761. Springer

  40. [48]

    Shao, H.; Yang, Y.; Lin, H.; Lin, L.; Chen, Y.; Yang, Q.; and Zhao, H. 2022. Rethinking controllable variational autoencoders. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 19250--19259

  41. [49]

    Shrestha, R.; Zou, Y.; Chen, Q.; Li, Z.; Xie, Y.; and Deng, S. 2024. FairRAG: Fair human generation via fair retrieval augmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 11996--12005

  42. [50]

    Tishby, N.; and Zaslavsky, N. 2015. Deep learning and the information bottleneck principle. In 2015 ieee information theory workshop (itw), 1--5. IEEE

  43. [51]

    Tommasi, T.; Patricia, N.; Caputo, B.; and Tuytelaars, T. 2017. A deeper look at dataset bias. Domain adaptation in computer vision applications, 37--55

  44. [52]

    B.; Valera, I.; Gomez-Rodriguez, M.; and Gummadi, K

    Zafar, M. B.; Valera, I.; Gomez-Rodriguez, M.; and Gummadi, K. P. 2019. Fairness constraints: A flexible approach for fair classification. Journal of Machine Learning Research, 20(75): 1--42

  45. [53]

    B.; Valera, I.; Rogriguez, M

    Zafar, M. B.; Valera, I.; Rogriguez, M. G.; and Gummadi, K. P. 2017. Fairness constraints: Mechanisms for fair classification. In Artificial intelligence and statistics, 962--970. PMLR

  46. [54]

    Zbontar, J.; Jing, L.; Misra, I.; LeCun, Y.; and Deny, S. 2021. Barlow twins: Self-supervised learning via redundancy reduction. In International conference on machine learning, 12310--12320. PMLR

  47. [55]

    Zhang, F.; He, Q.; Kuang, K.; Liu, J.; Chen, L.; Wu, C.; Xiao, J.; and Zhang, H. 2024. Distributionally Generative Augmentation for Fair Facial Attribute Classification. arXiv preprint arXiv:2403.06606

  48. [56]

    Zhang, F.; Kuang, K.; Chen, L.; Liu, Y.; Wu, C.; and Xiao, J. 2023. Fairness-aware Contrastive Learning with Partially Annotated Sensitive Attributes. In The Eleventh International Conference on Learning Representations

  49. [57]

    S.; Zhang, H

    Zhang, M.; Sohoni, N. S.; Zhang, H. R.; Finn, C.; and Re, C. 2022. Correct-N-Contrast: a Contrastive Approach for Improving Robustness to Spurious Correlations. In International Conference on Machine Learning, 26484--26516. PMLR

  50. [58]

    Zhang, Z.; Song, Y.; and Qi, H. 2017. Age progression/regression by conditional adversarial autoencoder. In Proceedings of the IEEE conference on computer vision and pattern recognition, 5810--5818

  51. [59]

    Zhu, Z.; Yao, Y.; Sun, J.; Li, H.; and Liu, Y. 2023. Weak proxies are sufficient and preferable for fairness with missing sensitive attributes. In International Conference on Machine Learning, 43258--43288. PMLR

  52. [60]

    Zietlow, D.; Lohaus, M.; Balakrishnan, G.; Kleindessner, M.; Locatello, F.; Sch \"o lkopf, B.; and Russell, C. 2022. Leveling down in computer vision: Pareto inefficiencies in fair deep classifiers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog...

Pith tools

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