pith. sign in

arxiv: 2606.21107 · v1 · pith:LFUAWEP3new · submitted 2026-06-19 · 💻 cs.LG · cs.CR· stat.ME

Enhancing Differentially Private Mechanisms via Empirical Bayes

Minwoo Kim , Junyong Park , Sungkyu Jung This is my paper

Pith reviewed 2026-06-26 14:44 UTC · model grok-4.3

classification 💻 cs.LG cs.CRstat.ME
keywords differential privacyempirical BayesGaussian mechanismdenoisinghistogram releaseprincipal component analysislinear regressionmean squared error
0
0 comments X

The pith

Empirical Bayes estimation reduces mean squared error in Gaussian differential privacy outputs using only the noisy results as input.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that empirical Bayes estimation can denoise the output of the additive Gaussian mechanism for differential privacy. This reduces mean squared error without needing knowledge of the data distribution or extra privacy budget. The approach is applied to histogram release, principal component analysis, and linear regression, where it often outperforms other private algorithms in experiments. A sympathetic reader would care because it offers a simple post-processing step that improves utility for existing DP mechanisms.

Core claim

The authors establish that the empirical Bayes approach can reduce the mean-squared error solely by taking the output of the Gaussian mechanism as input. This denoising improves the utility of differential privacy mechanisms while preserving the original privacy guarantee, and numerical studies confirm gains on histogram release, PCA, and linear regression compared to existing methods.

What carries the argument

Empirical Bayes estimation applied directly to the noisy output of the Gaussian mechanism to estimate a prior and produce a denoised estimate.

If this is right

  • Histogram release under differential privacy achieves lower error at the same privacy level.
  • Private principal component analysis benefits from reduced error via this post-processing.
  • Linear regression under the Gaussian mechanism sees improved accuracy.
  • The method remains computationally simple and applies across multiple statistical tasks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The technique might extend to other additive noise mechanisms where a prior can be estimated from noisy observations.
  • It could serve as a default post-processing layer in differential privacy toolkits for Gaussian-based releases.
  • In settings with very high dimensions, careful validation of the prior estimation step would be needed to maintain gains.
  • Combinations with other utility-enhancing techniques could compound the error reductions observed here.

Load-bearing premise

An empirical Bayes prior can be reliably estimated from the noisy Gaussian mechanism output alone without knowledge of the underlying data distribution or additional private information.

What would settle it

If applying the empirical Bayes procedure to Gaussian-noisy outputs from known test distributions yields no reduction in mean squared error relative to the raw noisy values, the central claim would fail.

Figures

Figures reproduced from arXiv: 2606.21107 by Junyong Park, Minwoo Kim, Sungkyu Jung.

Figure 1
Figure 1. Figure 1: Average population L2-costs of private histogram release under Model I. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Average population L2-costs of private histogram release under Model II. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Simulation results for private PCA under a Gaussian distribution. [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Simulation results for private PCA under a multivariate [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Average population L2-costs of private histogram release under Model I with high-privacy setting. # of bins = 25 # of bins = 50 # of bins = 100 eps = 1 eps = 2 500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000 5 6 7 8 5 6 7 8 N log L2 loss Method Non−private Laplace Laplace−NPMLE [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Average population L2-costs of private histogram release under Model I with low￾privacy setting. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Average population L2-costs of private histogram release under Model II with high-privacy setting. # of bins = 25 # of bins = 50 # of bins = 100 eps = 1 eps = 2 500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000 6 7 8 6 7 8 N log L2 loss Method Non−private Laplace Laplace−NPMLE [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Average population L2-costs of private histogram release under Model II with low-privacy setting. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
read the original abstract

Differential privacy (DP) has become the gold standard for ensuring the privacy protection of machine learning and statistical algorithms in recent decades. A plethora of algorithms and methods have been developed to enhance the utility of DP algorithms while maintaining the same level of DP. However, these are often overly complex or computationally ineffective. We propose a novel approach focusing on denoising the output of the simple additive Gaussian mechanism by adopting the idea of \textit{empirical Bayes estimation}. We highlight that the empirical Bayes approach can reduce the mean-squared error solely by taking the output of the Gaussian mechanism as input. Our numerical studies show that this simple yet powerful approach can be applied to improve upon various statistical problems, including histogram release, principal component analysis, and linear regression, often outperforming existing private algorithms.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes a post-processing step that applies empirical Bayes estimation to denoise the output of the additive Gaussian mechanism, claiming that this reduces mean-squared error using only the noisy output as input and without knowledge of the underlying data distribution. Numerical studies are presented for histogram release, principal component analysis, and linear regression, asserting outperformance relative to existing private algorithms.

Significance. If the empirical Bayes denoising step reliably improves MSE without additional privacy expenditure or external information, the approach would supply a lightweight, mechanism-agnostic utility boost applicable to many existing DP releases. The simplicity of the method is a potential strength relative to more complex DP enhancements.

major comments (2)
  1. [Numerical studies] Abstract and numerical-studies section: the central claim that an empirical Bayes prior estimated solely from the Gaussian-mechanism output yields lower MSE rests on the stability of deconvolution from the observed marginal. The manuscript provides no diagnostics (e.g., estimation-error curves, sensitivity to sample size, or explicit failure regimes) that would confirm the deconvolution remains well-behaved in the finite-sample regimes used for the histogram, PCA, and regression examples.
  2. [Method] Method description: the weakest assumption—that a reliable shrinkage rule can be recovered from noisy observations alone—is load-bearing for the MSE-reduction claim. No theoretical bound or consistency argument is supplied showing that the empirical-Bayes estimator converges to a useful denoiser when the true prior is misspecified or the number of pooled statistics is moderate.
minor comments (1)
  1. [Abstract] The abstract states that numerical studies demonstrate improvement but does not reference specific tables or figures; cross-references should be added once the experimental details are expanded.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. We address each major comment below, indicating where revisions will be made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Numerical studies] Abstract and numerical-studies section: the central claim that an empirical Bayes prior estimated solely from the Gaussian-mechanism output yields lower MSE rests on the stability of deconvolution from the observed marginal. The manuscript provides no diagnostics (e.g., estimation-error curves, sensitivity to sample size, or explicit failure regimes) that would confirm the deconvolution remains well-behaved in the finite-sample regimes used for the histogram, PCA, and regression examples.

    Authors: We agree that explicit diagnostics would better support the claims. In the revised manuscript we will add sensitivity plots with respect to the number of pooled statistics, estimation-error curves for the recovered prior, and a brief discussion of observed failure regimes in the numerical-studies section. revision: yes

  2. Referee: [Method] Method description: the weakest assumption—that a reliable shrinkage rule can be recovered from noisy observations alone—is load-bearing for the MSE-reduction claim. No theoretical bound or consistency argument is supplied showing that the empirical-Bayes estimator converges to a useful denoiser when the true prior is misspecified or the number of pooled statistics is moderate.

    Authors: The contribution is empirical; the manuscript demonstrates MSE reduction via numerical studies rather than supplying consistency theorems. We will revise the method section and conclusion to state this scope limitation explicitly and to reference relevant nonparametric empirical-Bayes literature on practical performance under moderate sample sizes. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical Bayes post-processing of Gaussian outputs is independent of inputs

full rationale

The paper applies standard empirical Bayes shrinkage to the outputs of the additive Gaussian mechanism, claiming MSE reduction from the noisy observations alone. No quoted derivation reduces a claimed result to a fitted parameter or self-referential definition by construction. No load-bearing self-citations, uniqueness theorems from the same authors, or ansatzes smuggled via prior work are evident in the provided text. The central claim rests on the well-known property that empirical Bayes estimators can improve MSE for multiple observations under known noise variance, which is externally verifiable and not forced by the paper's own inputs. This is the most common honest finding for a paper whose method is a direct application of an established technique.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review performed on abstract only; no explicit free parameters, axioms, or invented entities are stated in the provided text.

axioms (1)
  • standard math Gaussian mechanism satisfies differential privacy under standard assumptions
    Implicit background from DP literature invoked by the proposal.

pith-pipeline@v0.9.1-grok · 5658 in / 1055 out tokens · 16314 ms · 2026-06-26T14:44:23.295442+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

177 extracted references · 10 canonical work pages

  1. [1]

    Generalized iterative

    ElSalamouny, Ehab and Palamidessi, Catuscia , booktitle=. Generalized iterative. 2020 , volume=

    2020

  2. [2]

    and Cerna, Selene and Palamidessi, Catuscia

    Arcolezi, H \'e ber H. and Cerna, Selene and Palamidessi, Catuscia. On the utility gain of iterative Bayesian update for locally differentially private mechanisms. Data and Applications Security and Privacy XXXVII. 2023

    2023

  3. [3]

    Proceedings of the twenty-sixth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems , pages=

    Privacy, accuracy, and consistency too: a holistic solution to contingency table release , author=. Proceedings of the twenty-sixth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems , pages=

  4. [4]

    IEEE INFOCOM 2019-IEEE Conference on Computer Communications , pages=

    Calibrate: Frequency estimation and heavy hitter identification with local differential privacy via incorporating prior knowledge , author=. IEEE INFOCOM 2019-IEEE Conference on Computer Communications , pages=. 2019 , organization=

    2019

  5. [5]

    Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages=

    Maximum likelihood postprocessing for differential privacy under consistency constraints , author=. Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages=

  6. [6]

    An improved stochastic gradient descent algorithm based on R

    Cheng, XianFu and Yao, YanQing and Zhang, Liying and Liu, Ao and Li, Zhoujun , journal=. An improved stochastic gradient descent algorithm based on R. 2022 , publisher=

    2022

  7. [7]

    2009 Ninth IEEE International Conference on Data Mining , pages=

    Accurate estimation of the degree distribution of private networks , author=. 2009 Ninth IEEE International Conference on Data Mining , pages=. 2009 , organization=

    2009

  8. [8]

    Proceedings of the VLDB Endowment , volume=

    Boosting the accuracy of differentially private histograms through consistency , author=. Proceedings of the VLDB Endowment , volume=. 2010 , publisher=

    2010

  9. [9]

    Advances in Neural Information Processing Systems , volume=

    Probabilistic inference and differential privacy , author=. Advances in Neural Information Processing Systems , volume=

  10. [10]

    The Annals of Statistics , volume=

    Inference using noisy degrees: Differentially private -model and synthetic graphs , author=. The Annals of Statistics , volume=. 2016 , publisher=

    2016

  11. [11]

    Proceedings of the forty-fifth annual ACM symposium on Theory of computing , pages=

    The geometry of differential privacy: the sparse and approximate cases , author=. Proceedings of the forty-fifth annual ACM symposium on Theory of computing , pages=

  12. [12]

    Proceedings of the 34th International Conference on Machine Learning , pages =

    Differentially Private Learning of Undirected Graphical Models Using Collective Graphical Models , author =. Proceedings of the 34th International Conference on Machine Learning , pages =. 2017 , editor =

    2017

  13. [13]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Bias and variance of post-processing in differential privacy , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  14. [14]

    Advances in Neural Information Processing Systems , volume=

    Post-processing private synthetic data for improving utility on selected measures , author=. Advances in Neural Information Processing Systems , volume=

  15. [15]

    NDSS , year=

    Locally differentially private frequency estimation with consistency , author=. NDSS , year=

  16. [16]

    arXiv preprint arXiv:2007.11524 , year=

    Improving deep learning with differential privacy using gradient encoding and denoising , author=. arXiv preprint arXiv:2007.11524 , year=

    arXiv 2007

  17. [17]

    Wang, Wenxiao and Wang, Tianhao and Wang, Lun and Luo, Nanqing and Zhou, Pan and Song, Dawn and Jia, Ruoxi , journal=

  18. [18]

    2023 IEEE Smart World Congress (SWC) , pages=

    A deep learning-based data usability enhancement scheme for differential privacy , author=. 2023 IEEE Smart World Congress (SWC) , pages=. 2023 , organization=

    2023

  19. [19]

    2025 , journal=

    Robust and Differentially Private PCA for non-Gaussian data , author=. 2025 , journal=

    2025

  20. [20]

    1999 , publisher=

    A Wavelet Tour of Signal Processing , author=. 1999 , publisher=

    1999

  21. [21]

    Journal of the american statistical association , volume=

    Adapting to unknown smoothness via wavelet shrinkage , author=. Journal of the american statistical association , volume=. 1995 , publisher=

    1995

  22. [22]

    arXiv preprint arXiv:2402.13531 , year=

    Private gradient descent for linear regression: Tighter error bounds and instance-specific uncertainty estimation , author=. arXiv preprint arXiv:2402.13531 , year=

    arXiv

  23. [23]

    2024 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) , pages=

    Improved differentially private regression via gradient boosting , author=. 2024 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) , pages=. 2024 , organization=

    2024

  24. [24]

    2025 , note =

    ks: Kernel Smoothing , author =. 2025 , note =

    2025

  25. [25]

    The Eleventh International Conference on Learning Representations , year=

    Easy Differentially Private Linear Regression , author=. The Eleventh International Conference on Learning Representations , year=

  26. [26]

    Proceedings of the VLDB Endowment , month = jul, pages =

    Zhang, Jun and Zhang, Zhenjie and Xiao, Xiaokui and Yang, Yin and Winslett, Marianne , title =. Proceedings of the VLDB Endowment , month = jul, pages =. 2012 , issue_date =

    2012

  27. [27]

    The Noisy Power Method: A Meta Algorithm with Applications , volume =

    Hardt, Moritz and Price, Eric , booktitle =. The Noisy Power Method: A Meta Algorithm with Applications , volume =

  28. [28]

    Mathematical and Scientific Machine Learning , pages=

    Stochastic and private nonconvex outlier-robust PCAs , author=. Mathematical and Scientific Machine Learning , pages=. 2022 , organization=

    2022

  29. [29]

    Proceedings of the 40th International Conference on Machine Learning , pages =

    From Robustness to Privacy and Back , author =. Proceedings of the 40th International Conference on Machine Learning , pages =. 2023 , editor =

    2023

  30. [30]

    Differentially Private M-Estimators , url =

    Lei, Jing , booktitle =. Differentially Private M-Estimators , url =

  31. [31]

    Empirical Bayes: From

    Ignatiadis, Nikolaos and Sen, Bodhisattva , year =. Empirical Bayes: From

  32. [32]

    NSF-CBMS Regional Conference Series in Probability and Statistics , pages=

    Mixture Models: Theory, Geometry and Applications , author=. NSF-CBMS Regional Conference Series in Probability and Statistics , pages=. 1995 , organization=

    1995

  33. [33]

    Uncertainty in Artificial Intelligence (UAI-18) , year=

    Revisiting differentially private linear regression: optimal and adaptive prediction & estimation in unbounded domain , author=. Uncertainty in Artificial Intelligence (UAI-18) , year=

  34. [34]

    A Statistical Viewpoint on Differential Privacy: Hypothesis Testing, Representation, and Blackwell's Theorem

    Su, Weijie J. A Statistical Viewpoint on Differential Privacy: Hypothesis Testing, Representation, and Blackwell's Theorem. Annual Review of Statistics and Its Application. 2025. doi:https://doi.org/10.1146/annurev-statistics-112723-034158

    work page doi:10.1146/annurev-statistics-112723-034158 2025

  35. [35]

    Journal of Computational and Graphical Statistics , volume =

    Youngseok Kim and Peter Carbonetto and Matthew Stephens and Mihai Anitescu , title =. Journal of Computational and Graphical Statistics , volume =. 2020 , publisher =

    2020

  36. [36]

    Johnstone and Bernard W

    Iain M. Johnstone and Bernard W. Silverman , title =. The Annals of Statistics , number =

  37. [37]

    Journal of Machine Learning Research , year =

    Zhengrong Xing and Peter Carbonetto and Matthew Stephens , title =. Journal of Machine Learning Research , year =

  38. [38]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Multivariate, heteroscedastic empirical bayes via nonparametric maximum likelihood , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2025 , publisher=

    2025

  39. [39]

    The Complexity of Differential Privacy

    Vadhan, Salil. The Complexity of Differential Privacy. Tutorials on the Foundations of Cryptography: Dedicated to Oded Goldreich. 2017. doi:10.1007/978-3-319-57048-8_7

    work page doi:10.1007/978-3-319-57048-8_7 2017

  40. [40]

    Brendan Murphy and Adrian E

    Luca Scrucca and Chris Fraley and T. Brendan Murphy and Adrian E. Raftery , publisher =. Model-Based Clustering, Classification, and Density Estimation Using. doi:10.1201/9781003277965 , year =

    work page doi:10.1201/9781003277965

  41. [41]

    Machine learning and the

    Efron, Bradley , journal=. Machine learning and the. 2024 , publisher=

    2024

  42. [42]

    Proceedings of the 2nd Berkeley Symposium on Mathematical Statistics and Probability , volume=

    Asymptotically subminimax solutions of compound statistical decision problems , author=. Proceedings of the 2nd Berkeley Symposium on Mathematical Statistics and Probability , volume=. 1951 , organization=

    1951

  43. [43]

    An Empirical Bayes Approach to Statistics

    Robbins, Herbert E. An Empirical Bayes Approach to Statistics. Breakthroughs in Statistics: Foundations and Basic Theory. 1992. doi:10.1007/978-1-4612-0919-5_26

    work page doi:10.1007/978-1-4612-0919-5_26 1992

  44. [44]

    Journal of Statistical Software , volume=

    REBayes: an R package for empirical Bayes mixture methods , author=. Journal of Statistical Software , volume=

  45. [45]

    The Annals of Statistics , volume=

    General maximum likelihood empirical bayes estimation of normal means , author=. The Annals of Statistics , volume=. 2009 , publisher=

    2009

  46. [46]

    Journal of Statistical Software , volume=

    deconvolveR: A G-modeling program for deconvolution and empirical Bayes estimation , author=. Journal of Statistical Software , volume=

  47. [47]

    Journal of the American Statistical Association , volume=

    Convex optimization, shape constraints, compound decisions, and empirical Bayes rules , author=. Journal of the American Statistical Association , volume=. 2014 , publisher=

    2014

  48. [48]

    Nonparametric empirical

    Brown, Lawrence D and Greenshtein, Eitan , journal=. Nonparametric empirical. 2009 , volume =

    2009

  49. [49]

    2021 , publisher=

    Computer age statistical inference, student edition: algorithms, evidence, and data science , author=. 2021 , publisher=

    2021

  50. [50]

    Journal of the American Statistical Association , volume=

    Tweedie’s formula and selection bias , author=. Journal of the American Statistical Association , volume=. 2011 , publisher=

    2011

  51. [51]

    IEEE transactions on information theory , volume=

    De-noising by soft-thresholding , author=. IEEE transactions on information theory , volume=. 1995 , publisher=

    1995

  52. [52]

    Electronic journal of Statistics , year=

    Differentially private multivariate statistics with an application to contingency table analysis , author=. Electronic journal of Statistics , year=

  53. [53]

    Differential

    Dwork, Cynthia and Lei, Jing , title =. Proceedings of the Forty-First Annual ACM Symposium on Theory of Computing , pages =. 2009 , isbn =. doi:10.1145/1536414.1536466 , abstract =

    work page doi:10.1145/1536414.1536466 2009

  54. [54]

    Proceedings of The 26th International Conference on Artificial Intelligence and Statistics , pages =

    Generalized PTR: User-Friendly Recipes for Data-Adaptive Algorithms with Differential Privacy , author =. Proceedings of The 26th International Conference on Artificial Intelligence and Statistics , pages =. 2023 , editor =

    2023

  55. [55]

    The American Statistician , volume=

    An introduction to empirical Bayes data analysis , author=. The American Statistician , volume=. 1985 , publisher=

    1985

  56. [56]

    Proceedings of the Thirty-Ninth Annual ACM Symposium on Theory of Computing , pages =

    Nissim, Kobbi and Raskhodnikova, Sofya and Smith, Adam , title =. Proceedings of the Thirty-Ninth Annual ACM Symposium on Theory of Computing , pages =. 2007 , isbn =. doi:10.1145/1250790.1250803 , abstract =

    work page doi:10.1145/1250790.1250803 2007

  57. [57]

    On lines and planes of closest fit to systems of points in space , author=

    LIII. On lines and planes of closest fit to systems of points in space , author=. The London, Edinburgh, and Dublin philosophical magazine and journal of science , volume=. 1901 , publisher=

    1901

  58. [58]

    Robust and differentially private mean estimation , url =

    Liu, Xiyang and Kong, Weihao and Kakade, Sham and Oh, Sewoong , booktitle =. Robust and differentially private mean estimation , url =

  59. [59]

    arXiv preprint , pages =

    Targeted backdoor attacks on deep learning systems using data poisoning , author=. arXiv preprint , pages =

  60. [60]

    Proceedings of the 32nd International Conference on Machine Learning , pages =

    Is Feature Selection Secure against Training Data Poisoning? , author =. Proceedings of the 32nd International Conference on Machine Learning , pages =. 2015 , editor =

    2015

  61. [61]

    Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

    The US Census Bureau adopts differential privacy , author=. Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

  62. [62]

    Nature , volume=

    Genes mirror geography within Europe , author=. Nature , volume=. 2008 , publisher=

    2008

  63. [63]

    Journal of Statistical Planning and Inference , volume=

    Sign and rank covariance matrices , author=. Journal of Statistical Planning and Inference , volume=. 2000 , publisher=

    2000

  64. [64]

    Symmetric

    Fang, Kai-Tai and Kotz, Samuel Kotz and Ng, Kai Wang , year=. Symmetric

  65. [65]

    Statistics and Computing , volume=

    Generalized spherical principal component analysis , author=. Statistics and Computing , volume=. 2024 , publisher=

    2024

  66. [66]

    The Annals of Statistics , volume=

    Rate-optimal perturbation bounds for singular subspaces with applications to high-dimensional statistics , author=. The Annals of Statistics , volume=

  67. [67]

    Journal of Computational and Graphical Statistics , number=

    Distributed Learning for Principal Eigenspaces without Moment Constraints , author=. Journal of Computational and Graphical Statistics , number=. 2024 , publisher=

    2024

  68. [68]

    Conference on Learning Theory , pages=

    Fast rates for empirical risk minimization of strict saddle problems , author=. Conference on Learning Theory , pages=. 2017 , organization=

    2017

  69. [69]

    Statistics & probability letters , volume=

    Some robust estimates of principal components , author=. Statistics & probability letters , volume=. 1999 , publisher=

    1999

  70. [70]

    Van Erven, Tim and Harremos, Peter , journal=. R. 2014 , publisher=

    2014

  71. [71]

    Calibrating Noise to Sensitivity in Private Data Analysis

    Dwork, Cynthia and McSherry, Frank and Nissim, Kobbi and Smith, Adam. Calibrating Noise to Sensitivity in Private Data Analysis. Theory of Cryptography. 2006

    2006

  72. [72]

    Advances in Neural Information Processing Systems , volume=

    Differentially Private Covariance Revisited , author=. Advances in Neural Information Processing Systems , volume=

  73. [73]

    Algorithmic Learning Theory , pages=

    Old techniques in differentially private linear regression , author=. Algorithmic Learning Theory , pages=. 2019 , organization=

    2019

  74. [74]

    Advances in neural information processing systems , volume=

    Instance-optimality in differential privacy via approximate inverse sensitivity mechanisms , author=. Advances in neural information processing systems , volume=

  75. [75]

    Conference on Learning Theory , pages=

    An improved gap-dependency analysis of the noisy power method , author=. Conference on Learning Theory , pages=. 2016 , organization=

    2016

  76. [76]

    The Thirty Sixth Annual Conference on Learning Theory , pages=

    Fast, sample-efficient, affine-invariant private mean and covariance estimation for subgaussian distributions , author=. The Thirty Sixth Annual Conference on Learning Theory , pages=. 2023 , organization=

    2023

  77. [77]

    Proceedings of the 55th Annual ACM Symposium on Theory of Computing , pages=

    Robustness implies privacy in statistical estimation , author=. Proceedings of the 55th Annual ACM Symposium on Theory of Computing , pages=

  78. [78]

    Proceedings of the 55th Annual ACM Symposium on Theory of Computing , pages=

    Privately estimating a Gaussian: Efficient, robust, and optimal , author=. Proceedings of the 55th Annual ACM Symposium on Theory of Computing , pages=

  79. [79]

    Advances in neural information processing systems , volume=

    Differentially private robust low-rank approximation , author=. Advances in neural information processing systems , volume=

  80. [80]

    2018 , publisher=

    Han, Fang and Liu, Han , journal=. 2018 , publisher=

    2018

Showing first 80 references.