Advances and open problems in federated learning

P Kairouz, HB McMahan, B Avent… - … and trends® in …, 2021 - nowpublishers.com
Federated learning (FL) is a machine learning setting where many clients (eg, mobile
devices or whole organizations) collaboratively train a model under the orchestration of a …

Why is public pretraining necessary for private model training?

A Ganesh, M Haghifam, M Nasr, S Oh… - International …, 2023 - proceedings.mlr.press
In the privacy-utility tradeoff of a model trained on benchmark language and vision tasks,
remarkable improvements have been widely reported when the model is pretrained on …

Efficient mean estimation with pure differential privacy via a sum-of-squares exponential mechanism

SB Hopkins, G Kamath, M Majid - Proceedings of the 54th Annual ACM …, 2022 - dl.acm.org
We give the first polynomial-time algorithm to estimate the mean of ad-variate probability
distribution with bounded covariance from Õ (d) independent samples subject to pure …

Private distribution learning with public data: The view from sample compression

S Ben-David, A Bie, CL Canonne… - Advances in …, 2023 - proceedings.neurips.cc
We study the problem of private distribution learning with access to public data. In this setup,
which we refer to as* public-private learning*, the learner is given public and private …

Private estimation with public data

A Bie, G Kamath, V Singhal - Advances in neural …, 2022 - proceedings.neurips.cc
We initiate the study of differentially private (DP) estimation with access to a small amount of
public data. For private estimation of $ d $-dimensional Gaussians, we assume that the …

A private and computationally-efficient estimator for unbounded gaussians

G Kamath, A Mouzakis, V Singhal… - … on Learning Theory, 2022 - proceedings.mlr.press
We give the first polynomial-time, polynomial-sample, differentially private estimator for the
mean and covariance of an arbitrary Gaussian distribution $ N (\mu,\Sigma) $ in $\R^ d $. All …

New lower bounds for private estimation and a generalized fingerprinting lemma

G Kamath, A Mouzakis… - Advances in neural …, 2022 - proceedings.neurips.cc
We prove new lower bounds for statistical estimation tasks under the constraint of
$(\varepsilon,\delta) $-differential privacy. First, we provide tight lower bounds for private …

Decision tree for locally private estimation with public data

Y Ma, H Zhang, Y Cai, H Yang - Advances in Neural …, 2024 - proceedings.neurips.cc
We propose conducting locally differentially private (LDP) estimation with the aid of a small
amount of public data to enhance the performance of private estimation. Specifically, we …

A polynomial time, pure differentially private estimator for binary product distributions

V Singhal - International Conference on Algorithmic …, 2024 - proceedings.mlr.press
We present the first $\varepsilon $-differentially private, computationally efficient algorithm
that estimates the means of product distributions over $\{0, 1\}^ d $ accurately in total …

Leveraging public representations for private transfer learning

P Thaker, A Setlur, ZS Wu, V Smith - arxiv preprint arxiv:2312.15551, 2023 - arxiv.org
Motivated by the recent empirical success of incorporating public data into differentially
private learning, we theoretically investigate how a shared representation learned from …