Abstract:Many problems in machine learning and other fields can be (re)formulated as linearly constrained separable convex programs. In most of the cases, there are multiple blocks of variables. However, the traditional alternating direction method (ADM) and its linearized version (LADM, obtained by linearizing the quadratic penalty term) are for the two-block case and cannot be naively generalized to solve the multi-block case. So there is great demand on extending the ADM based methods for the multi-block case. In this paper, we propose LADM with parallel splitting and adaptive penalty (LADMPSAP) to solve multi-block separable convex programs efficiently. When all the component objective functions have bounded subgradients, we obtain convergence results that are stronger than those of ADM and LADM, e.g., allowing the penalty parameter to be unbounded and proving the sufficient and necessary conditions for global convergence. We further propose a simple optimality measure and reveal the convergence rate of LADMPSAP in an ergodic sense. For programs with extra convex set constraints, with refined parameter estimation we devise a practical version of LADMPSAP for faster convergence. Finally, we generalize LADMPSAP to handle programs with more difficult objective functions by linearizing part of the objective function as well. LADMPSAP is particularly suitable for sparse representation and low-rank recovery problems because its subproblems have closed form solutions and the sparsity and low-rankness of the iterates can be preserved during the iteration. It is also highly parallelizable and hence fits for parallel or distributed computing. Numerical experiments testify to the advantages of LADMPSAP in speed and numerical accuracy.

A Sequential ADMM Algorithm to Find Sparse LCP Solutions Using a L2−l1 Regularization Technique with …

A Sequential ADMM Algorithm to Find Sparse LCP Solutions Using a L2-L1 Regularization Technique with Application in Bimatrix Game.

Nonnegative Iterative Reweighted Method for Sparse Linear Complementarity Problem

Linearized Alternating Direction Method of Multipliers for Constrained Nonconvex Regularized Optimization.

A Half Thresholding Projection Algorithm for Sparse Solutions of LCPs.

Linearized Alternating Direction Method with Parallel Splitting and Adaptive Penalty for Separable Convex Programs in Machine Learning

Two methods for finding a sparse solution of the linear complementarity problem with z-matrix

Adaptive Linearized Alternating Direction Method of Multipliers for Non-Convex Compositely Regularized Optimization Problems

A Nonconvex ADMM for a Class of Sparse Inverse Semidefinite Quadratic Programming Problems

A Lagrangian Dual Based Approach to Sparse Linear Programming

The Linearized Alternating Direction Method of Multipliers for Sparse Group LAD Model.

Linearized Alternating Direction Method with Adaptive Penalty for Low-Rank Representation

Optimal linearized symmetric ADMM for multi-block separable convex programming

Linearized Alternating Direction Method with Adaptive Penalty for Low-Rank Representation.

An Inertial Proximal Partially Symmetric ADMM-based Algorithm for Linearly Constrained Multi-Block Nonconvex Optimization Problems with Applications.

A Constrained Optimization Approach for Lcp

A Bregman-Style Improved ADMM and its Linearized Version in the Nonconvex Setting: Convergence and Rate Analyses

A new penalized least absolute deviation model for high dimensional sparse linear regression and an efficient sequential linear programming algorithm

Square-Root Lasso with Nonconvex Regularization: an ADMM Approach

On Partial Smoothness, Activity Identification and Faster Algorithms of L1 over L2 Minimization

A Generalization of Linearized Alternating Direction Method of Multipliers for Solving Two-Block Separable Convex Programming