Efficient and robust high-dimensional sparse logistic regression via nonlinear primal-dual hybrid gradient algorithms

TL;DR

This paper introduces a nonlinear primal-dual algorithm for high-dimensional sparse logistic regression that is more efficient and robust, especially suited for big data sets with many predictor variables.

Contribution

The paper presents a novel nonlinear primal-dual algorithm with provable convergence that significantly improves computational complexity for large-scale sparse logistic regression.

Findings

Achieves $O(T(m,n) ext{log}(1/\epsilon))$ complexity, outperforming traditional methods.

Handles big data sets efficiently with scalable matrix-vector operations.

Provides reliable numerical results in high-dimensional variable selection tasks.

Abstract

Logistic regression is a widely used statistical model to describe the relationship between a binary response variable and predictor variables in data sets. It is often used in machine learning to identify important predictor variables. This task, variable selection, typically amounts to fitting a logistic regression model regularized by a convex combination of $\ell_1$ and $\ell_{2}^{2}$ penalties. Since modern big data sets can contain hundreds of thousands to billions of predictor variables, variable selection methods depend on efficient and robust optimization algorithms to perform well. State-of-the-art algorithms for variable selection, however, were not traditionally designed to handle big data sets; they either scale poorly in size or are prone to produce unreliable numerical results. It therefore remains challenging to perform variable selection on big data sets without access to adequate and costly computational resources. In this paper, we propose a nonlinear primal-dual algorithm that addresses these shortcomings. Specifically, we propose an iterative algorithm that provably computes a solution to a logistic regression problem regularized by an elastic net penalty in $O(T(m,n)\log(1/\epsilon))$ operations, where $\epsilon \in (0,1)$ denotes the tolerance and $T(m,n)$ denotes the number of arithmetic operations required to perform matrix-vector multiplication on a data set with $m$ samples each comprising $n$ features. This result improves on the known complexity bound of $O(\min(m^2n,mn^2)\log(1/\epsilon))$ for first-order optimization methods such as the classic primal-dual hybrid gradient or forward-backward splitting methods.