Solving Lasso: Extended ADMM is more efficient than ADMM

Abstract

The least absolute shrinkage and selection operator (Lasso) has become very popular and attractive approach for regularization and variable selection for high-dimensional data in machine learning. In this paper, we present an extended alternating direction method of multipliers (ADMM) for solving the Lasso. The extended ADMM is global convergent, and all the subproblems can easily get the solutions. It can also be implemented in distributed manner, which is beneficial for storage and computation requirement. Numerical experiments demonstrated that the extended ADMM outperforms other popular algorithms.