Properties and Iterative Methods for the Elastic Net with ℓp-Norm Errors

Liling Wei,Hong-Kun Xu
DOI: https://doi.org/10.24193/fpt-ro.2020.2.57
2020-01-01
Fixed Point Theory
Abstract:The p-elastic net (p-EN) with 1 < p < infinity is introduced to recover a sparse signal x is an element of R-n from m (< n) linear measurements with noise. The p-EN, which extends the elastic net of Zou and Hastie [23] and was implicitly suggested by Tropp [16], amounts to minimizing the objective function (1/p)parallel to Ax + b parallel to(p)(p) + lambda parallel to x parallel to(1) + (mu/2)parallel to x parallel to(2)(2) over x is an element of R-n, where A is the measurement matrix, b is the observation, and lambda > 0, mu > 0 are regularization parameters. Some basic geometric properties of the p-EN such as how the solution curve of the minimization depends on the parameters lambda and mu are investigated. Moreover, iterative algorithms such as the proximal-gradient algorithm and the Frank-Wolfe algorithm are studied for solving the p-EN.
What problem does this paper attempt to address?