Improved $ell_{p}$ Regression via Iteratively Reweighted Least Squares

2025-10-02

AI Summary

This paper addresses the ℓₚ regression problem to bridge the performance gap between theoretical algorithms and practical applications. We propose a novel primal-dual optimization framework based on iterative reweighted least squares (IRLS), which exploits the invariance of the dual objective to derive simple, closed-form update rules—bypassing expensive subproblem solves. The method retains the optimal O(log(1/ε)) iteration complexity while requiring only lightweight matrix operations per iteration, drastically reducing per-step computational overhead. Rigorous theoretical analysis establishes global convergence and complexity optimality. Empirical evaluation demonstrates superior accuracy, speed, and robustness compared to the Adil–Peng–Sachdeva IRLS algorithm and mainstream solvers including MATLAB/CVX. The key innovation lies in the tight integration of primal-dual structure with IRLS, achieving both theoretical rigor and engineering practicality.

Technology Category

Machine Learning: OptimizationSearch and Optimization: Non-convex OptimizationComputer Vision: Learning & Optimization for CV

Application Category

Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphsSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingSystems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments

Abstract

We introduce fast algorithms for solving $ell_{p}$ regression problems using the iteratively reweighted least squares (IRLS) method. Our approach achieves state-of-the-art iteration complexity, outperforming the IRLS algorithm by Adil-Peng-Sachdeva (NeurIPS 2019) and matching the theoretical bounds established by the complex algorithm of Adil-Kyng-Peng-Sachdeva (SODA 2019, J. ACM 2024) via a simpler lightweight iterative scheme. This bridges the existing gap between theoretical and practical algorithms for $ell_{p}$ regression. Our algorithms depart from prior approaches, using a primal-dual framework, in which the update rule can be naturally derived from an invariant maintained for the dual objective. Empirically, we show that our algorithms significantly outperform both the IRLS algorithm by Adil-Peng-Sachdeva and MATLAB/CVX implementations.

Problem

• Developing fast algorithms for ℓp regression problems
• Achieving state-of-the-art iteration complexity via IRLS
• Bridging theoretical-practical gap with primal-dual framework

Innovation

• Using iteratively reweighted least squares method
• Employing primal-dual framework for update rule
• Achieving state-of-the-art iteration complexity