- Rating
- 3
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper studies fair representation learning when the sensitive attribute is continuous. It argues that averaging the fairness violation over the sensitive attribute can hide large violations in low-density ranges. To address this issue, the authors propose supIPM, the largest ReLU-IPM between the representation distribution given and its marginal, over all . They give a kernel estimator and a training method named supFR, together with a uniform convergence rate and a bound on the worst-case disparity of H"older smooth prediction heads. Experiments on tabular and graph datasets show much lower representation-level disparity than seven baselines with similar accuracy.
Critical strengths
-
The paper is well motivated. Figure 1 and the simulation in Figure 2 show how averaging can hide a violation in a rare range of .
-
The proposed method is supported with theoretical guarantees. The bound in Theorem 1 is uniform over encoders, so it applies to the learned encoder, and Corollary 2 extends it to every head in a H"older ball.
-
The method is validated in various datasets, and the results look promising.
Critical weaknesses
-
The theory does not cover the cases that motivate the paper. Theorem 1 and Corollary 2 hold only on the interior region with , and the constants depend on the lower bound of the density of . But the violations in the motivation examples are in low-density regions.
-
The main evaluation metric is close to what supFR optimizes. Both take a worst case over of a function of a linear score of the representation. On measures that supFR does not train against (like kernel GDP, binned GDP in Appendix D.5), it's only comparable to the baselines. So part of the large gain may come from matching the training and test criteria.
Questions for authors
-
Could you clarify the values of and used in the experiments? How much of the observed lies outside the region covered by Theorem 1, and how does the constant change as becomes small?
-
Could you repeat the simulation in Sec 5.1 with a more extreme value than the current setting = 0.8? That can help validate if the method works in the motivating case even where the theory does not apply.
-
Could you clarify if the performance gain is not from the matching of training and test criteria? The authors may consider reporting a representation-level disparity that supFR does not train against.
Overall, the paper studies an important problem, and proposed method is well motivated. If the authors could clarify the gap between the theory and the motivating cases used in the paper, and whether the performance gain is from training w.r.t. the testing criteria, I'd be happy to further increase my score.
4 comments
Question 1
- For , we min-max scale to and take the supremum over , that is, . The table below gives the share of observations in the training and test splits outside this range.
Share of observations outside .
| Adult | Crime | ACSIncome | |
|---|---|---|---|
| excluded (%) | 1.2 | 9.5 | 0.6 |
- The share is larger on Crime because every excluded observation there takes the smallest or the largest observed value of , which min-max scaling maps to zero and one.
- The constant appears only in the theory, so neither the algorithm nor the evaluation requires its value. The constant of Theorem 1 grows like by Eq. (51) in Appendix A, so the bound becomes looser as becomes small, and a larger sample is needed to attain the same error bound.
Question 2
- We thank the reviewer for this suggestion, which led to the new Appendix C.4.1 with . We also ran the simulation with , even closer to the boundary. In both cases the conclusion of Figure 2 holds, and supFR attains a smaller representation-level disparity than FREM, as the table below shows.
Representation-level disparity at matched prediction performance.
| supFR | FREM | |
|---|---|---|
| test at least 0.945 | ||
| 0.0018 | 0.0068 | |
| 0.0030 | 0.0075 | |
| 0.0026 | 0.0084 |
Question 3
- We thank the reviewer for this suggestion. The new Figure 14 of Appendix D.5 reports the kernel GDP of the worst head, that is, the downstream head attaining the representation-level disparity, which supFR does not train against. supFR attains the smallest value among all methods on the three datasets, as the table below shows.
Kernel GDP of the worst head at matched prediction performance.
| supFR | FREM | |
|---|---|---|
| Adult, accuracy 0.842 | 0.0009 | 0.0096 |
| Crime, 0.855 | 0.0026 | 0.0104 |
| ACSIncome, accuracy 0.784 | 0.0009 | 0.0077 |
-
We think that this gain comes from the ReLU discriminator class, which controls the disparity of every Hölder smooth head by Corollary 2 rather than that of a single trained head.
-
We thank the reviewer again for the constructive comments. We hope that our responses have addressed the reviewer's concerns, and we would be happy to answer any further questions.
Thank you for the detailed rebuttal and the revised paper. After reading the authors' rebuttal, I still have two concerns remaining:
-
For Weakness 1, Proposition 7 is a meaningful addition that addresses my concern of the boundary. However, the guarantee is weaker in the low-density setting which motivates the paper. The bound depends on a single c_S for the whole range, so one sparse region weakens the guarantee everywhere, not just near that region. So I agree the guarantee holds theoretically, but it is not quite clear how informative the guarantee is in practice.
-
For Weakness 2, the addition of Reg-GDP$\infty$ didn't address my question of training w.r.t. the evaluation metric, as Reg-GDP$\infty$ also optimizes the worst case. Beating it shows supFR is a better worst-case method, but it doesn’t show that supFR’s advantage would survive a test it wasn’t trained for. In fact, the new results in Appendix D seem to suggest the opposite. When unfairness is measured with the adversarial linear head (i.e., the kind supFR trains against), supFR is significantly better, but when the same unfairness measure is applied to the ordinary trained prediction head, supFR is only comparable to the baselines. Since only the head changed, the big advantage seems to be from matching the evaluator to the training criterion. Also, the added results in Table 5 are a bit surprising to me. Reg-GDP$\infty$ directly considers the worst-case, but is worse than Reg-GDP on Adult and ACSIncome dataset. This might be because the Reg-GDP$\infty$ method is under-tuned.
We thank the reviewer for the follow-up comments. Our responses to the two remaining concerns are given below.
Question 1
- Theorem 1 and Proposition 7 show that the empirical supIPM converges uniformly to the supIPM and give the rate of convergence. By Corollary 2, a small empirical supIPM then guarantees a small disparity at every sensitive value, including an of low density.
- The lower bound appears only in the constant and not in the rate. Thus the bound still goes to zero as grows when is small, and only a larger sample is needed to attain the same error bound.
- In the simulation of Appendix C.4.1, the density of is with , so that . We also ran the simulation with , for which . For both values of , supFR attains a smaller representation-level disparity than FREM, as the table below shows. The unfair model is trained without the fairness penalty, that is, with .
Representation-level disparity at matched prediction performance.
| Unfair | supFR | FREM | ||
|---|---|---|---|---|
| test at least 0.945 | ||||
| , (Figure 6) | 0.1 | 0.1025 | 0.0030 | 0.0075 |
| , | 0.01 | 0.0508 | 0.0018 | 0.0056 |
- Relative to the unfair model, the disparity of supFR is for and for . Thus supFR is only slightly less effective for the smaller .
Question 2
- To examine whether the gain comes from matching the evaluator to the training criterion, we change the fairness measure of the evaluation. The most unfair linear head is now chosen to maximize the kernel GDP, a measure that supFR does not train against. supFR again attains the smallest value among all methods on the three datasets, as the table below shows.
Kernel GDP of the most unfair linear head at matched prediction performance.
| supFR | FREM | |
|---|---|---|
| Adult, accuracy 0.842 | 0.0009 | 0.0101 |
| Crime, 0.855 | 0.0033 | 0.0131 |
| ACSIncome, accuracy 0.784 | 0.0010 | 0.0086 |
- The linear heads used in the evaluation are Hölder smooth, so by Corollary 2 supFR controls their disparity at every . The kernel GDP averages this disparity over , so it is small as well.
- Regarding Table 5, we thank the reviewer for pointing out that Reg-GDP$_\infty$ may be under-tuned. Following this comment, we retrained Reg-GDP$_\infty$ on Adult and ACSIncome with a finer grid of and updated Table 5. Its prediction-level disparity is now smaller than that of Reg-GDP on Adult and ACSIncome.
Disparities at matched prediction performance.
| supFR | Reg-GDP | Reg-GDP$_\infty$ | |
|---|---|---|---|
| Adult, accuracy 0.842 | |||
| prediction-level | 0.0350 | 0.0639 | 0.0618 |
| representation-level | 0.0016 | 0.0343 | 0.0348 |
| ACSIncome, accuracy 0.784 | |||
| prediction-level | 0.0345 | 0.0487 | 0.0378 |
| representation-level | 0.0018 | 0.0206 | 0.0159 |
Final remark
Our aim is to learn a fair representation that controls the unfairness arising at a sensitive value of low density. supFR is therefore designed to reduce the worst-case disparity of downstream heads. We hope that our responses resolve the reviewer's concerns. We would be grateful if the reviewer could let us know of any further questions.
We thank the reviewer for the careful review and for pointing out the gap between the theory on the interior region and the low-density and boundary cases that motivate the paper. This comment led to a new guarantee over the entire sensitive domain. Revisions made in response to the reviewer's comments are marked in blue in the manuscript.
Weakness 1
Uniform convergence of the empirical supIPM with bandwidth γn.
Weakness 2
Disparities at matched prediction performance.