Gradient Estimation Based Electromagnetism-like Algorithm for Sparse Optimization Problems

Abstract

In this study we equip the electromagnetism-like algorithm (EM method) proposed by Birbil Fang and Sheu with the ability of handling sparsity By solving the convex minimization sub-problems we obtain the LASSO gradient estimates We recover the sparsity information with these gradient estimates and use these gradient estimates as the search direction in the mirror descent algorithm From our numerical testings retrieving the sparsity information by LASSO gradient estimation as well as incorporating with the mirror descent algorithm does save the computational time and improve the solution quality

Date of Award	2021
Original language	English
Supervisor	Ruey-Lin Sheu (Supervisor)