We study objectives ℱ d combining a quadratic data-fidelity and an ℓ 0 regularization. Data d are generated using a full-rank M × N matrix A with N>M. Our main results are listed below. Minimizers û of ℱ d are strict if and only if length(support(û)) ≤ M and the submatrix of A whose columns are indexed by support(û) is full rank. Their continuity in data is derived. Global minimizers are always strict. We adopt a weak assumption on A and show that it holds with probability one. Data read d = Aü where length(support(ü))≤ M - 1 and the submatrix whose columns are indexed by support(ü) is full rank. Among all strict (local) minimizers of ℱ d with support shorter than M-1, the exact solution û = ü is the unique vector that cancels the residual. The claim is independent of the regularization parameter. This û = ü is usually a strict local minimizer where ℱ d does not reach its global minimum. Global minimization of ℱ d can then prevent the recovery of ü. A numerical example (A is 5 x 10) illustrates our main results. © 2012 Springer-Verlag.
CITATION STYLE
Nikolova, M. (2012). Should we search for a global minimizer of least squares regularized with an ℓ 0 penalty to get the exact solution of an under determined linear system? In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 6667 LNCS, pp. 508–519). https://doi.org/10.1007/978-3-642-24785-9_43
Mendeley helps you to discover research relevant for your work.