Faster Algorithms for User-Level Private Stochastic Convex Optimization

TL;DR

This paper introduces faster, more practical user-level differential privacy algorithms for stochastic convex optimization, achieving optimal excess risk with significantly reduced computational complexity and fewer assumptions.

Contribution

The authors develop novel user-level DP algorithms for SCO that are faster, require fewer assumptions, and achieve optimal excess risk compared to prior methods.

Findings

Linear-time algorithm with state-of-the-art excess risk under mild smoothness.

Optimal excess risk achieved with approximately (mn)^{9/8} gradient computations.

Optimal excess risk for non-smooth losses in n^{11/8} m^{5/4} gradient computations.

Abstract

We study private stochastic convex optimization (SCO) under user-level differential privacy (DP) constraints. In this setting, there are $n$ users (e.g., cell phones), each possessing $m$ data items (e.g., text messages), and we need to protect the privacy of each user's entire collection of data items. Existing algorithms for user-level DP SCO are impractical in many large-scale machine learning scenarios because: (i) they make restrictive assumptions on the smoothness parameter of the loss function and require the number of users to grow polynomially with the dimension of the parameter space; or (ii) they are prohibitively slow, requiring at least $(mn)^{3/2}$ gradient computations for smooth losses and $(mn)^3$ computations for non-smooth losses. To address these limitations, we provide novel user-level DP algorithms with state-of-the-art excess risk and runtime guarantees, without stringent assumptions. First, we develop a linear-time algorithm with state-of-the-art excess risk (for a non-trivial linear-time algorithm) under a mild smoothness assumption. Our second algorithm applies to arbitrary smooth losses and achieves optimal excess risk in $\approx (mn)^{9/8}$ gradient computations. Third, for non-smooth loss functions, we obtain optimal excess risk in $n^{11/8} m^{5/4}$ gradient computations. Moreover, our algorithms do not require the number of users to grow polynomially with the dimension.