Pith. sign in

REVIEW 2 major objections 5 minor 60 references

Private Rate-Constrained Optimization with Applications to Fair Learning

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper introduces RaCO-DP, a differentially private training method that enforces arbitrary rate constraints—group fairness, false-negative limits, and more—by reducing their cost to a private histogram per mini-batch.

desk verdict New rate-constrained DP formulation with solid theory for the unclipped case, but the main theorem does not cover the clipped algorithm actually evaluated. read the letter →

arxiv 2505.22703 v2 pith:BCBRXTMS submitted 2025-05-28 cs.LG cs.CRstat.ML

Mohammad Yaghini , Tudor Cebere , Michael Menart , Aurélien Bellet , Nicolas Papernot This is my paper
classification cs.LGcs.CRstat.ML
keywords differentialprivacyrateconstraintsgroupfairnessstochasticgradientdescent-ascentLagrangiandualityprivatehistogramsnon-convexoptimizationfalsenegative
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 asks whether constrained machine learning can be made differentially private when the constraints depend on statistics of prediction rates across subpopulations—demographic parity, equality of odds, false-negative limits—rather than on per-example losses. The authors' answer is RaCO-DP, a stochastic gradient descent-ascent algorithm that solves the Lagrangian formulation of the constrained problem and whose only extra privacy cost is a private histogram of each mini-batch. They prove that, under Lipschitz and smoothness assumptions, the algorithm converges to an approximate stationary point of the min-max objective, with the approximation error set by the privacy budget. The reported experiments show RaCO-DP meeting specified fairness and false-negative targets while Pareto-dominating the previous state of the art in the fairness-utility trade-off. If the central claim holds, rate constraints are no longer an obstacle to private learning but a standard, privacy-compatible objective.

What carries the argument

The machinery is the generalized rate constraint form Gamma_j($\theta$) = sum_{I in I_j} sum_{k in [K]} alpha_{j,I,k} P_k(union_{i in I} D_i; $\theta$), where {D_1,...,D_Q} is a single public partition of the dataset and each constraint recombines these parts. This form guarantees that the only non-decomposable statistics are the counts used to normalize each rate, and those counts are exactly the entries of a private histogram H-$hat^{{(t)}}$ with L1-sensitivity 1. The optimization engine is SGDA on the Lagrangian: a projected gradient-ascent update on the dual variable $\lambda$ whose gradient is the private constraint violation, and a per-sample clipped-and-noised gradient descent update on $\theta$ that uses the same private histogram through post-processing. The proof machinery that carries the convergence argument is the linear structure of the dual update, which lets the analysis rely on an L-infinity bound on the dual gradient error instead of L2, and which yields a 1/$T^{{1/4}}$ rate rather than the 1/$T^{{1/6}}$ rate of the standard nonconvex-concave GDA analysis.

What would settle it

Run Algorithm 1 exactly as written—with the Line 8 gradient clipping and with the parameter choices of Theorem D.1—on a logistic-regression task with the strict false-negative constraint FNR=0 used in the paper's Figure 6, in the noise-free setting (sigma=0, b=infinity). If for a small clipping norm such as C=2 the iterates never approach the feasible set, as the paper's own figure shows, then the claimed convergence to an (alpha,alpha)-stationary point of Phi does not hold for the actual (clipped) algorithm; a complete argument would need a bound on the clipping bias.

Watch

Extended reading notes

Core claim

The paper's central claim is that arbitrary rate constraints can be folded into differentially private optimization without weakening the privacy guarantee. It defines generalized rate constraints as weighted sums of prediction rates over unions of parts of a fixed partition of the dataset, and observes that all quantities needed for both the primal and dual updates—constraint values and their per-sample gradients—can be read off a single histogram of softmax predictions over that partition. Because each data point touches exactly one part and its class probabilities sum to one, the histogram has L1-sensitivity at most one, so Laplace noise suffices to privatize it. The primal update then applies DP-SGD-style clipping and Gaussian noise to per-sample Lagrangian gradients, while the dual update uses the private histogram's post-processed values and therefore costs no additional privacy budget. The mathematical payload is a convergence analysis of SGDA with biased gradients: exploiting the linearity of the Lagrangian in the dual parameter, the analysis shows that with suitable parameters some iterate is an ($\alpha$, $\alpha$)-stationary point of the value function Phi($\theta$) = max_{$\lambda$ in Lambda} L($\theta$, $\lambda$), with $\alpha$ scaling like (d log(JKn/rho) log(n/delta)/(n eps))^{1/3} plus $K^{{1/4}}$-dependent terms, and with a faster rate in T than previously known for nonconvex-concave SGDA. The paper presents this as the first general differentially private treatment of rate-constrained optimization, with group fairness as a prominent application.

Load-bearing premise

The convergence guarantee is proved for the algorithm run without gradient clipping, whereas the algorithm as stated and as run clips every per-sample gradient; if the bias introduced by clipping is not controlled, the advertised guarantee does not cover the method actually used.

Editorial extensions

If this is right

  • Group fairness constraints such as demographic parity and equalized odds become compatible with standard (epsilon,delta)-differential privacy, with the fairness-related cost reduced to a private histogram per step rather than per-example decomposition.
  • Practitioners can specify the maximum allowed disparity or false-negative rate directly through the slack gamma and obtain it, instead of tuning an indirect penalty coefficient as in prior fairness methods.
  • The improved SGDA analysis—1/T^{1/4} convergence using the linear structure of the dual and an L-infinity dual-gradient error—applies to any nonconvex-linear min-max problem with bounded gradient errors, beyond rate constraints.
  • For a privacy budget (epsilon,delta), the optimal number of iterations balances optimization progress against noise, yielding the stated error scale alpha = O((d log(JKn/rho) log(n/delta)/(n eps))^{1/3} + K^{1/4}(log(n/delta) log(JKn/rho))^{1/4}/(n eps)^{1/4}).

Reading between the lines

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

  • If the clipping-norm bias shown in the paper's Figure 6 is the dominant obstacle, a natural testable extension is to let the clipping norm grow during training, which might reconcile the implemented algorithm with the no-clipping theory while preserving privacy accounting.
  • The histogram route suggests that other aggregate statistics that are bounded per example—such as calibration or coverage constraints—could be brought into DP optimization by the same mechanism, since their evaluation reduces to a low-sensitivity histogram.
  • Once clipping is included in the analysis as a fixed bias term, the effective stationarity error alpha would degrade by roughly the bias magnitude, implying a three-way trade-off between privacy, feasibility, and utility that the current bound does not display explicitly.
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

2 major / 5 minor

Summary. The paper introduces RaCO-DP, a differentially private stochastic gradient descent-ascent method for minimizing an empirical loss subject to rate constraints, with group fairness constraints as the main application. The method reformulates the constrained problem through a Lagrangian, uses a generalized rate-constraint structure based on a global partition of the dataset, and privately estimates a per-mini-batch histogram at each step; the histogram is then post-processed to compute both the primal per-sample gradients and the dual constraint gradients. The paper states a privacy guarantee (Theorem 4.1), a formal convergence analysis for non-convex objectives under bias in the gradient estimates (Appendix D, Theorem D.1 and Theorem D.5), and experiments on demographic parity and false-negative-rate constraints across several tabular benchmarks, reporting Pareto improvements over DP-FERMI and near-non-private performance.

Significance. If the claimed convergence result were established for the actual algorithm, this would be a meaningful contribution: RaCO-DP would be the first general DP framework for arbitrary rate constraints, not just fairness, with formal privacy and convergence guarantees and a favorable empirical privacy-utility-fairness trade-off. The paper also has several genuine strengths: the histogram-based mechanism is a clean way to avoid per-constraint privacy composition; the convergence analysis is self-contained, does not rely on fitted constants, and explicitly attempts to handle biased gradient estimates; and the limitations discussion is unusually honest, including a direct demonstration that clipping bias can break constraint satisfaction. The main reason the paper cannot be accepted in its current form is that the central convergence theorem is proven only for a no-clipping variant of the algorithm, while Algorithm 1, the experiments, and the abstract all refer to the clipping-based method.

major comments (2)
  1. [Appendix D.5, Lemma D.4; Section 4.2, Eq. (10)] The main-text convergence claim is stated for Algorithm 1, but the formal theorem in Appendix D.2, Theorem D.1, is explicitly for 'Algorithm 1 run without clipping,' while Algorithm 1 as defined clips every per-sample gradient at Line 8 and all reported experiments use clipping. The gradient-error bound in Lemma D.4 does not model the clipping operation at all, so the proof does not establish convergence to an approximate stationary point for the algorithm whose results are actually reported. This is not a purely cosmetic mismatch: Figure 6 and the Limitations section show that clipping bias alone, with sigma=0 and b=infinity, pushes the FNR constraint violation above 0.6 for C<=2 on Adult, i.e., far outside the feasible set, and the cited work of Koloskova et al. indicates that such bias does not vanish with vanishing step sizes. The paper should either prove convergence with clipping, explicitly bound the clipping bias and its effect on the stationarity and feasibility guarantees, or formally restrict the main theorem to the no-clipping variant and correspondingly qualify the abstract, the informal Theorem 5.2, and the experimental claims.
  2. [Appendix D.5, Lemma D.4; Section 4.2, Eq. (10)] The notation H in Lemma D.4 needs clarification. In Section 4, H^{(t)} is defined as the mini-batch histogram (Eq. (8)), and the per-sample gradient in Eq. (10) uses H^{(t)} as the mini-batch histogram, with sum over k of H_{i,k} equal to |B∩I|. In the proof of the primal gradient error, the first term in the regularizer difference sums over x in D while normalizing by sum_{i in I, k} H_{i,k}; if H is the mini-batch histogram, this term is not the full-batch gradient ∇θR(θ,λ). If H is instead intended to be a full-batch histogram for the purposes of the proof, that should be stated explicitly and consistently. As written, the proof does not transparently derive the claimed bound relating the full-batch gradient to the private mini-batch estimator, and this gap is load-bearing for Theorem D.1.
minor comments (5)
  1. [Section 5, before Theorem 5.2] The sentence 'This leads to the following result for Algorithm 1 run without clipping' should be incorporated into the theorem statement itself, so that the informal Theorem 5.2 cannot be read as applying to the clipping-based Algorithm 1 defined in Section 4.
  2. [Appendix D.5, Lemma D.4] Please define all symbols (H versus \widehat H, full-batch versus mini-batch histograms) before the proof, and make the decomposition of the primal regularizer error explicit: sampling error, private-histogram denominator error, and loss sampling error should be separated.
  3. [Section 6, Experimental setup] The statement 'we do not account for privacy loss from hyperparameter tuning' means that the reported ε values are not end-to-end privacy guarantees for the models selected via validation on the training data. This is standard in parts of the literature, but it should be stated more prominently and preferably discussed in the Limitations section.
  4. [Section 6, Baselines and experimental setup] The comparison with DP-FERMI uses results reported in prior work while reserving 15% of the training set for validation in the current setup. Please clarify how this difference in training-set size affects comparability, since the baseline numbers were generated under a different pipeline.
  5. [Throughout] There are several typos and notation inconsistencies, including 'Langrangian' in Section 4.1, inconsistent use of ε versus ϵ, and the informal theorem statement referring to 'Algorithm 1' rather than 'Algorithm 1 run without clipping.' These should be corrected in a revision.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the convergence and privacy analyses are derived from stated assumptions and noise-calibration bounds, with the acknowledged no-clipping caveat constituting a correctness gap rather than a circular step.

full rationale

RaCO-DP's central claim (Theorem 5.2 and its formal version, Theorem D.1) is a convergence and privacy statement derived from explicit assumptions: Lipschitzness and smoothness of h and ℓ, a compact dual set Λ, Poisson sampling, and calibrated Laplace/Gaussian noise scales. The proof chain is self-contained: Lemma D.4 bounds primal and dual gradient estimation errors using concentration inequalities (Lemmas C.1 and C.2), and Theorem D.5 converts these bounds into an approximate-stationarity guarantee for SGDA. No fitted constant, experiment, or empirical curve is used as an input to the proof, and the stated rate is parameter-free up to problem constants. The 'generalized rate constraints' formulation is an explicit generalization of the existing rate-constraint form (Eq. 5), not a renaming that presupposes the privacy or convergence result. Self-citations in the paper (e.g., [7], [8], [48]) appear in related-work or privacy-accounting contexts and are not load-bearing for the main theorem. The authors' own limitation section and Figure 6 document that the formal convergence theorem assumes Algorithm 1 is run without clipping, while the implemented algorithm clips per-sample gradients in Line 8; this is a scope/correctness gap between theory and practice, not circular reasoning, because Theorem D.1 does not assume its own conclusion and the gap is explicitly disclosed. No circular step satisfies the required standard of exhibiting a specific reduction of a claimed prediction to its inputs.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

No invented entities are proposed. The central theoretical result introduces no free constants beyond standard hyperparameters, but the empirical claim relies on validation-tuned hyperparameters, including a critical clipping norm that is not fully specified. The main axiomatic assumptions are smoothness and Lipschitzness, compactness of the dual set, and the fact that the convergence proof treats a no-clipping variant of Algorithm 1.

free parameters (6)
  • clipping norm C
    The paper shows in Figure 6 that C is critical and that small values bias iterates outside the feasible set, but Appendix E.2 does not report how C was selected or its final values.
  • Gaussian noise standard deviation σ
    Tuned in [3, 6] on validation accuracy and constraint satisfaction; it affects the privacy-utility trade-off.
  • Laplace noise scale b
    Tuned in [0.1, 0.5]; it controls the accuracy of the private histogram and the amount of noise added per step.
  • Primal and dual learning rates ηθ, ηλ
    Both are tuned in [1e-4, 0.1] across five seeds and affect whether the Lagrangian reaches a stationary point.
  • softmax temperature τ
    Tuned in [1, 10]; it approximates hard prediction rates and enters the Lipschitz and smoothness constants in the theory.
  • mini-batch size B and sampling rate r
    Tuned in [256, 1256]; it affects privacy amplification via Poisson subsampling and stochastic gradient noise.
assumptions (7)
  • ad hoc to paper Convergence is proven only for Algorithm 1 run without clipping
    Appendix D.2 Theorem D.1 states the algorithm is run without clipping, while Algorithm 1 Line 8 clips per-sample gradients; the main-text Theorem 5.2 does not repeat this qualification.
  • domain assumption Loss ℓ and classifier h are Lipschitz and smooth in θ
    Theorem D.1 and Lemma D.10 require finite Lipschitz and smoothness constants; logistic regression satisfies these, but neural networks are not globally smooth.
  • domain assumption The dual feasible set Λ is compact with bounded diameter and the Lagrangian is linear in λ
    Section 5 restricts Λ to a bounded set to make penalties finite; the convergence proof exploits linearity of L in the dual variable.
  • standard math Poisson subsampling and the composition theorems give privacy amplification
    Theorem 4.1 and D.3 rely on standard amplification by subsampling and advanced composition; these require the stated constraints on r, T, and per-round privacy loss.
  • domain assumption A known global partition of the dataset exists and the constraint weights α are public
    Section 3 assumes the constraints are written over disjoint subgroups; if sensitive attributes are not public or are noisy, the private histogram estimates constraint violations incorrectly.
  • domain assumption Mini-batch intersection sizes |B∩I| stay bounded away from zero with high probability
    Lemma D.2 lower-bounds these sizes; the primal and dual updates divide by them, and the gradient error analysis conditions on this event.
  • ad hoc to paper Privacy loss from hyperparameter tuning is not counted in the reported ε budget
    Section 6 states this standard practice; it means the formal privacy guarantee covers training only, not the selection of hyperparameters.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Private Rate-Constrained Optimization with Applications to Fair Learning." pith.science (2026). https://pith.science/paper/BCBRXTMS

@misc{pith2026250522703,
  author       = {Pith},
  title        = {Pith review of: Private Rate-Constrained Optimization with Applications to Fair Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BCBRXTMS}},
  note         = {Machine review of arXiv:2505.22703}
}
read the original abstract

Many problems in trustworthy ML can be expressed as constraints on prediction rates across subpopulations, including group fairness constraints (demographic parity, equalized odds, etc.). In this work, we study such constrained minimization problems under differential privacy (DP). Standard DP optimization techniques like DP-SGD rely on objectives that decompose over individual examples, enabling per-example gradient clipping and noise addition. Rate constraints, however, depend on aggregate statistics across groups, creating inter-sample dependencies that violate this decomposability. To address this, we develop RaCO-DP, a DP variant of Stochastic Gradient Descent-Ascent (SGDA) that solves the Lagrangian formulation of rate constraint problems. Through careful design, the extra privacy cost incurred by incorporating these constraints in our approach is limited to that of privately estimating a histogram over each mini-batch at every step. We prove the convergence of our algorithm through a novel analysis of SGDA that leverages the linear structure of the dual parameter. Empirical results show that our method Pareto-dominates existing private learning approaches under group fairness constraints and also achieves strong privacy-utility-fairness performance on neural networks.

Figures

Figures reproduced from arXiv: 2505.22703 by the authors.

Figure 1
Figure 1. Each rate constraint of the form (6) builds local datasets based on the global partition. A class-1 (class￾0) prediction is shown with a blue (red) square. Prediction rates P0, P1 are shown as fractions. As an example, let D1, D2, and D3 be the set of Hispanic, Black, and Caucasian individuals in the dataset, re￾spectively. Constraint Γ1 builds its local datasets as {{D1}, {D2 ∪ D3}}, i.e. {His￾panic, Non-Hispanic},… view at source ↗
Figure 2
Figure 2. (Left) Disparity-Error trade-off curves of DP fair training algorithms on Adult under demographic parity constraints. RaCO-DP Pareto dominates the SOTA method (DP-FERMI), closing the optimality gap with non-private (SGDA). (Right) RaCO-DP vs. Non-Private SGDA on ACSEmployment with 18 constraints, showing that RaCO-DP adapts to multiple sensitive groups. 6 Experimental Results To demonstrate RaCO-DP’s versatility, we… view at source ↗
Figure 3
Figure 3. Satisfiability on Adult. Trade-off between test error and constraint violation for different target values γ (dashed lines), av￾eraged over 20 runs. RaCO-DP achieves demographic parity con￾straint satisfaction. We use the numerical accountant from Doroshenko et al. [20] to track privacy loss during training. Based on hyperparameters affecting privacy (σ, b, and B) and for each privacy budget (ε, δ), we determine max… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Demographic Parity Constraint 30 [PITH_FULL_IMAGE:figures/full_fig_p030_4.png]
Figure 5
Figure 5. Figure 5: False Negative Rate Constraint 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 C 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Test constraint RaCO-DP [PITH_FULL_IMAGE:figures/full_fig_p031_5.png]
Figure 6
Figure 6. Figure 6: False Negative Rate-Constrained Classification on Adult. We show how the clipping norm C plays a critical role in satisfying a pessimistic constraint (γ = 0), even without noise related to differential privacy (σ = 0, b = ∞). 31 [PITH_FULL_IMAGE:figures/full_fig_p031_6.png]
Figure 7
Figure 7. Figure 7: Hard vs Soft Constraints on Adult. Trade-off between test error and demographic parity on Adult dataset. Dashed lines show target constraints, with soft (hollow circles) and hard (solid dots) constraint implementations achieving similar performance across different tar…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

60 extracted references · 57 canonical work pages

  1. [1]

    Deep Learning with Differential Privacy

    Martin Abadi et al. “Deep Learning with Differential Privacy”. In:Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security (CCS). 2016

    work page 2016

  2. [2]

    A Reductions Approach to Fair Classification

    Alekh Agarwal et al. “A Reductions Approach to Fair Classification”. In:Proceedings of the 35th International Conference on Machine Learning. Proceedings of Machine Learning Research. PMLR, 2018

    work page 2018

  3. [3]

    2022.url: https://www.kaggle.com/datasets/ alexteboul/heart-disease-health-indicators-dataset (visited on 05/16/2025)

    Alex Teboul.Heart Disease Health Indicators Dataset. 2022.url: https://www.kaggle.com/datasets/ alexteboul/heart-disease-health-indicators-dataset (visited on 05/16/2025)

    work page 2022

  4. [4]

    Differential Privacy Has Disparate Impact on Model Accuracy

    Eugene Bagdasaryan, Omid Poursaeed, and Vitaly Shmatikov. “Differential Privacy Has Disparate Impact on Model Accuracy”. In:Advances in Neural Information Processing Systems. 2019

    work page 2019

  5. [5]

    Privacy amplification by subsampling: Tight analyses via couplings and divergences

    Borja Balle, Gilles Barthe, and Marco Gaboardi. “Privacy amplification by subsampling: Tight analyses via couplings and divergences”. In:Advances in neural information processing systems31 (2018)

    work page 2018

  6. [6]

    MIT Press, 2023

    Solon Barocas, Moritz Hardt, and Arvind Narayanan.Fairness and Machine Learning: Limitations and Opportunities. MIT Press, 2023

    2023

  7. [7]

    Differentially Private Algorithms for the Stochastic Saddle Point Problem with Optimal Rates for the Strong Gap

    Raef Bassily, Cristóbal Guzmán, and Michael Menart. “Differentially Private Algorithms for the Stochastic Saddle Point Problem with Optimal Rates for the Strong Gap”. In:Proceedings of Thirty Sixth Conference on Learning Theory. Proceedings of Machine Learning Research. 2023

    work page 2023

  8. [8]

    Private Algorithmsfor Stochastic SaddlePoints and Variational Inequalities: Beyond Euclidean Geometry

    Raef Bassily, CristóbalA Guzmán, andMichael Menart. “Private Algorithmsfor Stochastic SaddlePoints and Variational Inequalities: Beyond Euclidean Geometry”. In:The Thirty-eighth Annual Conference on Neural Information Processing Systems. 2024

    work page 2024

Show all 60 references
  1. [9]

    Private Empirical Risk Minimization: Efficient Algorithms and Tight Error Bounds

    Raef Bassily, Adam D. Smith, and Abhradeep Thakurta. “Private Empirical Risk Minimization: Efficient Algorithms and Tight Error Bounds”. In:FOCS. 2014

    2014

  2. [10]

    UCI Machine Learning Repository

    Barry Becker and Ronny Kohavi.Adult. UCI Machine Learning Repository. 1996

    1996

  3. [11]

    Leonard Berrada et al.Unlocking Accuracy and Fairness in Differentially Private Image Classification. 2023

    2023

  4. [12]

    Optimal algorithms for differentially private stochastic monotone variational inequalities and saddle-point problems

    Digvijay Boob and Cristóbal Guzmán. “Optimal algorithms for differentially private stochastic monotone variational inequalities and saddle-point problems”. In:Mathematical Programming(2023)

    2023

  5. [13]

    Robust Optimization for Non-Convex Objectives

    Robert S. Chen et al. “Robust Optimization for Non-Convex Objectives”. In:NIPS. 2017

    2017

  6. [14]

    ScalableDP-SGD:Shufflingvs.poissonsubsampling

    LynnChuaetal.“ScalableDP-SGD:Shufflingvs.poissonsubsampling”.In: arXiv preprint arXiv:2411.04205 (2024)

    2024 arXiv

  7. [15]

    Optimization with Non-Differentiable Constraints with Applications to Fairness, Recall, Churn, and Other Goals

    Andrew Cotter et al. “Optimization with Non-Differentiable Constraints with Applications to Fairness, Recall, Churn, and Other Goals”. In:Journal of Machine Learning Research(2019)

    2019

  8. [16]

    Training Well-Generalizing Classifiers for Fairness Metrics and Other Data- Dependent Constraints

    Andrew Cotter et al. “Training Well-Generalizing Classifiers for Fairness Metrics and Other Data- Dependent Constraints”. In:Proceedings of the 36th International Conference on Machine Learning. International Conference on Machine Learning. PMLR, 2019

    2019

  9. [17]

    On the compatibility of privacy and fairness

    Rachel Cummings et al. “On the compatibility of privacy and fairness”. In:Adjunct publication of the 27th conference on user modeling, adaptation and personalization. 2019

    2019

  10. [18]

    Stochastic Model-Based Minimization of Weakly Convex Functions

    Damek Davis and Dmitriy Drusvyatskiy. “Stochastic Model-Based Minimization of Weakly Convex Functions”. In:SIAM Journal on Optimization29 (2019)

    2019

  11. [19]

    Retiring Adult: New Datasets for Fair Machine Learning

    Frances Ding et al. “Retiring Adult: New Datasets for Fair Machine Learning”. In:Advances in Neural Information Processing Systems34 (2021)

    2021

  12. [20]

    Connectthedots:Tighterdiscreteapproximationsofprivacylossdistributions

    VadymDoroshenkoetal.“Connectthedots:Tighterdiscreteapproximationsofprivacylossdistributions”. In: arXiv preprint arXiv:2207.04380(2022)

    2022 arXiv

  13. [21]

    The Algorithmic Foundations of Differential Privacy

    Cynthia Dwork and Aaron Roth. “The Algorithmic Foundations of Differential Privacy”. In:Foundations and Trends® in Theoretical Computer Science(2014)

    2014

  14. [22]

    Calibrating Noise to Sensitivity in Private Data Analysis

    Cynthia Dwork et al. “Calibrating Noise to Sensitivity in Private Data Analysis”. In:Theory of Cryptography. 2006. 11

    2006

  15. [23]

    Fairness through awareness

    Cynthia Dwork et al. “Fairness through awareness”. In:Proceedings of the 3rd innovations in theoretical computer science conference. 2012

    2012

  16. [24]

    Disparate Impact in Differential Privacy from Gradient Misalignment

    Maria S Esipova et al. “Disparate Impact in Differential Privacy from Gradient Misalignment”. In:The Eleventh International Conference on Learning Representations. 2022

    2022

  17. [25]

    Neither private nor fair: Impact of data imbalance on utility and fairness in differential privacy

    Tom Farrand et al. “Neither private nor fair: Impact of data imbalance on utility and fairness in differential privacy”. In:Proceedings of the 2020 Workshop on Privacy-Preserving Machine Learning in Practice. 2020

    2020

  18. [26]

    Deep Learning with Label Differential Privacy

    Badih Ghazi et al. “Deep Learning with Label Differential Privacy”. In:Advances in Neural Information Processing Systems. Vol. 34. 2021

    2021

  19. [27]

    Non-Discriminatory Machine Learning through Convex Fairness Criteria

    Naman Goel, Mohammad Yaghini, and Boi Faltings. “Non-Discriminatory Machine Learning through Convex Fairness Criteria”. In:Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society. 2018

    2018

  20. [28]

    Satisfying real-world goals with dataset constraints

    Gabriel Goh et al. “Satisfying real-world goals with dataset constraints”. In:Advances in neural information processing systems29 (2016)

    2016

  21. [29]

    Mirror Descent Algorithms with Nearly Dimension-Independent Rates for Differentially-Private Stochastic Saddle-Point Problems extended abstract

    Tomas Gonzalez, Cristobal Guzman, and Courtney Paquette. “Mirror Descent Algorithms with Nearly Dimension-Independent Rates for Differentially-Private Stochastic Saddle-Point Problems extended abstract”. In:Proceedings of Thirty Seventh Conference on Learning Theory. Proceedin...

    2024

  22. [30]

    Numerical Composition of Differential Privacy

    Sivakanth Gopi, Yin Tat Lee, and Lukas Wutschitz. “Numerical Composition of Differential Privacy”. In: Advances in Neural Information Processing Systems. Vol. 34. 2021

    2021

  23. [31]

    Stochastic-Differentially-Private-and-Fair-Learning

    Devansh Gupta. Stochastic-Differentially-Private-and-Fair-Learning. May 6, 2023. url: https : / / github.com/devanshgupta160/Stochastic- Differentially- Private- and- Fair- Learning (vis- ited on 09/27/2023)

    2023

  24. [32]

    Equality of opportunity in supervised learning

    Moritz Hardt, Eric Price, and Nati Srebro. “Equality of opportunity in supervised learning”. In:Advances in neural information processing systems29 (2016)

    2016

  25. [33]

    GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium

    Martin Heusel et al. “GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium”. In:Advances in Neural Information Processing Systems. Vol. 30. Curran Associates, Inc., 2017

    2017

  26. [34]

    Statlog (German Credit Data)

    Hans Hofmann. Statlog (German Credit Data). UCI Machine Learning Repository. 1994

    1994

  27. [35]

    Differentially private fair learning

    Matthew Jagielski et al. “Differentially private fair learning”. In:International Conference on Machine Learning. PMLR. 2019

    2019

  28. [36]

    The composition theorem for differential privacy

    Peter Kairouz, Sewoong Oh, and Pramod Viswanath. “The composition theorem for differential privacy”. In: International conference on machine learning. PMLR. 2015

    2015

  29. [37]

    What can we learn privately?

    Shiva Prasad Kasiviswanathan et al. “What can we learn privately?” In:SIAM Journal on Computing 40 (2011)

    2011

  30. [38]

    RevisitingGradientClipping:Stochastic bias and tight convergence guarantees

    AnastasiaKoloskova,HadrienHendrikx,andSebastianUStich.“RevisitingGradientClipping:Stochastic bias and tight convergence guarantees”. In:International Conference on Machine Learning. PMLR. 2023

    2023

  31. [39]

    What You See is What You Get: Principled Deep Learning via Distributional Generalization

    Bogdan Kulynych et al. “What You See is What You Get: Principled Deep Learning via Distributional Generalization”. In:Advances in Neural Information Processing Systems. 2022

    2022

  32. [40]

    Avoiding Pitfalls for Privacy Accounting of Subsampled Mechanisms under Composition

    Christian Janos Lebeda et al. “Avoiding Pitfalls for Privacy Accounting of Subsampled Mechanisms under Composition”. In:arXiv preprint arXiv:2405.20769(2024)

    2024 arXiv

  33. [41]

    On Gradient Descent Ascent for Nonconvex-Concave Minimax Problems

    Tianyi Lin, Chi Jin, and Michael Jordan. “On Gradient Descent Ascent for Nonconvex-Concave Minimax Problems”. In:Proceedings of the 37th International Conference on Machine Learning. International Conference on Machine Learning. PMLR, Nov. 21, 2020

    2020

  34. [42]

    Parkinsons

    Max Little. Parkinsons. UCI Machine Learning Repository. 2007

    2007

  35. [43]

    Stochastic Differentially Private and Fair Learning

    Andrew Lowy, Devansh Gupta, and Meisam Razaviyayn. “Stochastic Differentially Private and Fair Learning”. In:The Eleventh International Conference on Learning Representations. 2023. 12

    2023

  36. [44]

    Differential Privacy has Bounded Impact on Fairness in Classification

    Paul Mangold et al. “Differential Privacy has Bounded Impact on Fairness in Classification”. In:ICML. 2023

    2023

  37. [45]

    Performance analysis of cost-sensitive learning methods with application to imbalanced medical data

    Ibomoiye Domor Mienye and Yanxia Sun. “Performance analysis of cost-sensitive learning methods with application to imbalanced medical data”. In:Informatics in Medicine Unlocked(2021)

    2021

  38. [46]

    Fair learning with private demographic data

    Hussein Mozannar, Mesrob Ohannessian, and Nathan Srebro. “Fair learning with private demographic data”. In:International Conference on Machine Learning. PMLR. 2020

    2020

  39. [47]

    Robust Stochastic Approximation Approach to Stochastic Programming

    A. Nemirovski et al. “Robust Stochastic Approximation Approach to Stochastic Programming”. In: (2009)

    2009

  40. [48]

    Hyperparameter Tuning with Renyi Differential Privacy

    Nicolas Papernot and Thomas Steinke. “Hyperparameter Tuning with Renyi Differential Privacy”. In: International Conference on Learning Representations. 2022

    2022

  41. [49]

    Membership Inference Attacks against Machine Learning Models

    Reza Shokri et al. “Membership Inference Attacks against Machine Learning Models”. In:2017 IEEE Symposium on Security and Privacy (SP). 2017

    2017

  42. [50]

    Stochastic gradient descent with differ- entially private updates

    Shuang Song, Kamalika Chaudhuri, and Anand D. Sarwate. “Stochastic gradient descent with differ- entially private updates”. In:2013 IEEE Global Conference on Signal and Information Processing. 2013

    2013

  43. [51]

    Chasing Your Long Tails: Differentially Private Prediction in Health Care Settings

    Vinith M. Suriyakumar et al. “Chasing Your Long Tails: Differentially Private Prediction in Health Care Settings”. In:Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency. 2021

    2021

  44. [52]

    K-Anonymity: A Model for Protecting Privacy

    Latanya Sweeney. “K-Anonymity: A Model for Protecting Privacy”. In:Int. J. Uncertain. Fuzziness Knowl.-Based Syst.(Oct. 1, 2002)

    2002

  45. [53]

    Differentially Private Empirical Risk Minimization under the Fairness Lens

    Cuong Tran, My Dinh, and Ferdinando Fioretto. “Differentially Private Empirical Risk Minimization under the Fairness Lens”. In:Advances in Neural Information Processing Systems. Vol. 34. 2021

    2021

  46. [54]

    On the Fairness Impacts of Private Ensembles Models

    Cuong Tran and Ferdinando Fioretto. “On the Fairness Impacts of Private Ensembles Models”. In: Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence. 2023

    2023

  47. [55]

    SF-PATE: Scalable, Fair, and Private Aggregation of Teacher Ensembles

    Cuong Tran et al. “SF-PATE: Scalable, Fair, and Private Aggregation of Teacher Ensembles”. In: Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, IJCAI-23. International Joint Conferences on Artificial Intelligence Organization, 2023

    2023

  48. [56]

    Exact Privacy Analysis of the Gaussian Sparse Histogram Mechanism

    Arjun Wilkins et al. “Exact Privacy Analysis of the Gaussian Sparse Histogram Mechanism”. In:Journal of Privacy and Confidentiality14.1 (2024)

    2024

  49. [57]

    Differentially Private SGDA for Minimax Problems

    Zhenhuan Yang et al. “Differentially Private SGDA for Minimax Problems”. In:The 38th Conference on Uncertainty in Artificial Intelligence. 2022

    2022

  50. [58]

    Default of Credit Card Clients

    I-Cheng Yeh. Default of Credit Card Clients. UCI Machine Learning Repository. 2009

    2009

  51. [59]

    Fairness Constraints: Mechanisms for Fair Classification

    Muhammad Bilal Zafar et al. “Fairness Constraints: Mechanisms for Fair Classification”. In:Proceedings of the 20th International Conference on Artificial Intelligence and Statistics. 2017

    2017

  52. [60]

    Bring Your Own Algorithm for Optimal Differentially Private Stochastic Minimax Optimization

    Liang Zhang et al. “Bring Your Own Algorithm for Optimal Differentially Private Stochastic Minimax Optimization”. In:Advances in Neural Information Processing Systems. Vol. 35. Curran Associates, Inc., 2022. 13 A Application to Other Rate Constraints A.1 Fairness Constraints F...

    2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.