Regression under demographic parity constraints via unlabeled post-processing

We address the problem of performing regression while ensuring demographic parity, even without access to sensitive attributes during inference. We present a general-purpose post-processing algorithm that, using accurate estimates of the regression function and a sensitive attribute predictor, generates predictions that meet the demographic parity constraint. Our method involves discretization and stochastic minimization of a smooth convex function. It is suitable for online post-processing and multi-class classification tasks only involving unlabeled data for the post-processing. Unlike prior methods, our approach is fully theory-driven. We require precise control over the gradient norm of the convex function, and thus, we rely on more advanced techniques than standard stochastic gradient descent. Our algorithm is backed by finite-sample analysis and post-processing bounds, with experimental results validating our theoretical findings.

Paper

Similar papers

Peer review

Reviewer qMMa7/10 · confidence 4/52024-06-25

Summary

This paper considers post-processing regressors to satisfy the group fairness criterion of demographic parity, in particular, under the attribute unaware setting. 1. The authors begin by analyzing the constrained optimization problem (\*) for fair post-processing, and showing in lemma 3.1 that the solution, i.e., the post-processed regressor, can be represented and parameterized by the dual variables of (\*); and provided that the base models are accurate and (\*) is solved to optimiality, the resulting regressor will be optimally fair. 2. Then the authors discussed optimization and statistical aspects of (\*). Specifically, the convergence rate is analyzed, as well as the conditions for non-perfect solutions to satisfy fairness, namely that the "gradient mapping" of (\*) needs to be lipschitz. For this reason, the authors recommended using the SGD3 algorithm for optimizing (\*). 3. The paper closes with empirical evaluations of the proposed algorithm.

Strengths

1. To my knowledge, this is the first paper that studies fair post-processing regressors in the attribute unaware setting. 2. The primal-dual analysis that leads to the representation result in lemma 3.1 is interesting. Because the proposed procedure involves discretization, the formulation of (\*) is based on the support of the regressor's output space. This is different from somewhat similar works for the classification setting [1, 2] where the optimization problem is based on the scores of the training examples (without discretization). 3. The theoretical analysis is thorough, hence the proposed procedure is well-supported, including the choice of SGD3. [1] https://arxiv.org/pdf/2310.05725 [2] https://arxiv.org/pdf/2405.04025

Weaknesses

1. Regarding lipschitzness of $F$. The reviewer feels that some discussions are curt. The authors introduced the "gradient mapping" $G_\alpha$ in eq. (8), with a hyperparameter $\alpha$. It is not mentioned: how $\alpha$ should be chosen, in theory or in practice; how exactly SGD3 controls the norm of $G_\alpha$ (lines 237-239) in the main body. At least a brief discussion should be included in the main body for the latter, since theorem 5.1 would depend on whether SGD3 can provably reduce this quantity. 2. Regarding the use of SGD3, an ablation study would have been appreciated at illustrating the importance of using SGD3 over other optimizers; is it absolutely necessary? Also, could the authors discuss potential limitations with SGD3? 3. The authors mentioned that there are several hyperparameters associated with SGD3 (line 236), and with discretization (i.e., the number of bins). But it does not seem to be discussed how these are chosen for the experiments in section 6, and whether mis-specification of these hyperparameters would impact performance. [3] https://arxiv.org/pdf/2006.07286

Questions

- Does the analysis rely on properties unique to the demographic parity fairness criterion? If not, is there a path to extend to other criteria (potentially taking inspirations from [1, 2])?

Rating

7

Confidence

4

Soundness

3

Presentation

3

Contribution

3

Limitations

- See weaknesses. - Minor, but the proposed procedure requires discretizing the support, which is a limitation; even though the reviewer is aware that, in practice, discretization helps with generalization hence gives better performing models compared to non-parametric methods [3].

Reviewer RqJN7/10 · confidence 4/52024-07-10

Summary

This paper proposes an algorithm that takes in an fitted regression function and a sensitive attribute predictor and output a prediction function satisfying the demographic parity constraint. It designs a smooth convex objective function with discretization to solve for a prediction function with small risk and controlled violation of the demographic parity constraint. Stochastic minimization techniques are applied to solve the proposed optimization problem and recover the statistical rate $1/\sqrt{T}$.

Strengths

The proposed algorithm is supported by theoretical analysis and error bounds. The proofs are well-organized and clear to read. The algorithm deploys suitable stochastic minimization techniques to achieve statistical guarantees. Moreover, based on the experiment results, the proposed algorithm is much more computationally efficient than existing methods.

Weaknesses

Typos: * falls withing -> falls within (page 2) * statisitcal properties -> statistical properties (page 5) * out approach -> our approach (page 7) * phenomenons -> phenomena (page 7) * outperfomce -> outperform (page 8)

Questions

In practice, how do we pick algorithm parameters such as L, B and \beta?

Rating

7

Confidence

4

Soundness

4

Presentation

3

Contribution

3

Limitations

Yes.

Reviewer 3wik5/10 · confidence 3/52024-07-11

Summary

This paper presents a post-processing algorithm designed to enforce demographic parity in regression tasks without access to sensitive attributes during inference.

Strengths

The solution is versatile solution for enforcing demographic parity as it can be applied on different optimizers. The paper provides a rigorous theoretical foundation to quantify the risk. The fairness problem is relevant.

Weaknesses

While the algorithm presented in the paper appears promising, its evaluation is hampered by the presentation style, which tends to obscure clarity. Technical details such as the reliance on discretization and sophisticated methods for controlling the gradient norm are emphasized in the abstract, which could be streamlined to focus on broader impacts and significance instead. The paper contains numerous remarks that disrupt the flow of discussion; for instance, Remark 2.1 seems tangential and could be relegated to an appendix or omitted entirely. Other generalizations and discussions, such as those in Remark 2.2 and the paragraph on line 304, should be consolidated and presented at the end of the paper to maintain focus. The 'abuses notation' in Remark 3.1 could be resolved by providing clear definitions Comparisons with other literature are inconsistently integrated within the text, appearing at disparate locations such as lines 56, 228, and 334, which could be better organized to aid in comparative analysis and enhance readability. There is excessive use of sub-titles or mini-sections. Some of these sections are notably brief (line 176, for example), where the short content under each title does not justify the need for a separate heading.

Questions

The paper focuses on demographic parity as a fairness metric. Are there impacts on other fairness metrics like Equalized Odds and Equal Opportunity when using the proposed algorithm? Does it improve these metrics, or could it potentially worsen them? Regarding the risks plot in Figure 1, could you clarify its purpose given there are no comparative benchmarks provided? How should the unfairness score presented in the plot be interpreted? Reference [1] addresses a similar topic with a minimax approach. Is this minimax result applicable or relevant to the methodology used in your paper? In Algorithm 1, "DP" is mentioned in the name. Could you specify what "DP" stands for? The algorithm "fairname" appears in line 305 without a prior definition. Could you explain what this term means within the context of your study? [1] Fukuchi, Kazuto, and Jun Sakuma. "Demographic parity constrained minimax optimal regression under linear model." Advances in Neural Information Processing Systems 36 (2024).

Rating

5

Confidence

3

Soundness

1

Presentation

2

Contribution

2

Limitations

Not applicable.

Reviewer qMMa2024-08-11

The reviewer thanks the authors for the response. Could the authors also (briefly) comment on how the number of bins is chosen (in the experiments), and how it affects performance (theoretically)?

Authorsrebuttal2024-08-12

We appreciate the reviewer's feedback. In Lemma B.1 (Appendix B, line 518), we show that discretization increases the risk by $4B/L + 1/L^2 + \log(2L+1)/\beta$. Additionally, as demonstrated in the proof of Theorem 5.1 (Appendix D, line 615), setting $L = \sqrt{T}$ achieves a risk rate of $1/\sqrt{T}$. Therefore, we use $L = \sqrt{T}$ in our experiments, resulting in $2\sqrt{T} + 1$ bins, where T is the number of unlabeled samples. We also note that using more bins wouldn't improve statistical performance but would complicate the optimization due to higher dimensionality. Hence, we adhere to the theoretically recommended setting in practice.

Reviewer qMMa2024-08-13

The reviewer thanks the authors for the response, and think the readers would appreciate if these points are highlighted in the main body. The reviewer has raised their score.

Reviewer 3wik2024-08-13

Thank you for the the response, which has addressed most of my concerns. I believe the authors will improve the readability in the final version. I am no longer opposed to the acceptance of this paper and will adjust my score accordingly.

Reviewer RqJN2024-08-13

I thank the authors for their elaborate rebuttal. This does address my concerns and questions.

Program Chairsdecision2024-09-25

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC