Under low-rank ground truth, matrix RIP, and small noise relative to the penalty, the unique LASSO solution has rank approximately bounded by the ground truth, implying linear convergence of low-rank projected proximal gradient and global optimality of all second-order critical points in the non-con
Global convergence of sub-gradie nt method for robust matrix recovery: Small initialization, noisy measurements, and over-param eterization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Low solution rank of the matrix LASSO under RIP with consequences for rank-constrained algorithms
Under low-rank ground truth, matrix RIP, and small noise relative to the penalty, the unique LASSO solution has rank approximately bounded by the ground truth, implying linear convergence of low-rank projected proximal gradient and global optimality of all second-order critical points in the non-con