Improved Sample Complexity for Private Nonsmooth Nonconvex Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kornowski, Guy, Liu, Daogao, Talwar, Kunal
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909641988898816
author Kornowski, Guy
Liu, Daogao
Talwar, Kunal
author_facet Kornowski, Guy
Liu, Daogao
Talwar, Kunal
contents We study differentially private (DP) optimization algorithms for stochastic and empirical objectives which are neither smooth nor convex, and propose methods that return a Goldstein-stationary point with sample complexity bounds that improve on existing works. We start by providing a single-pass $(ε,δ)$-DP algorithm that returns an $(α,β)$-stationary point as long as the dataset is of size $\widetildeΩ(\sqrt{d}/αβ^{3}+d/εαβ^{2})$, which is $Ω(\sqrt{d})$ times smaller than the algorithm of Zhang et al. [2024] for this task, where $d$ is the dimension. We then provide a multi-pass polynomial time algorithm which further improves the sample complexity to $\widetildeΩ\left(d/β^2+d^{3/4}/εα^{1/2}β^{3/2}\right)$, by designing a sample efficient ERM algorithm, and proving that Goldstein-stationary points generalize from the empirical loss to the population loss.
format Preprint
id arxiv_https___arxiv_org_abs_2410_05880
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Improved Sample Complexity for Private Nonsmooth Nonconvex Optimization
Kornowski, Guy
Liu, Daogao
Talwar, Kunal
Machine Learning
Cryptography and Security
Optimization and Control
We study differentially private (DP) optimization algorithms for stochastic and empirical objectives which are neither smooth nor convex, and propose methods that return a Goldstein-stationary point with sample complexity bounds that improve on existing works. We start by providing a single-pass $(ε,δ)$-DP algorithm that returns an $(α,β)$-stationary point as long as the dataset is of size $\widetildeΩ(\sqrt{d}/αβ^{3}+d/εαβ^{2})$, which is $Ω(\sqrt{d})$ times smaller than the algorithm of Zhang et al. [2024] for this task, where $d$ is the dimension. We then provide a multi-pass polynomial time algorithm which further improves the sample complexity to $\widetildeΩ\left(d/β^2+d^{3/4}/εα^{1/2}β^{3/2}\right)$, by designing a sample efficient ERM algorithm, and proving that Goldstein-stationary points generalize from the empirical loss to the population loss.
title Improved Sample Complexity for Private Nonsmooth Nonconvex Optimization
topic Machine Learning
Cryptography and Security
Optimization and Control
url https://arxiv.org/abs/2410.05880