Recent insights in $\ell_1$ solvers

During this talk, an overview is given on our work on norm-one solvers as part of the DNOISE project. Gilles will explain the ins and outs of our iterative thresholding solver based on log cooling while Felix will present the work of Michael Friedlander "A Newton root-finding algorithms for large-scale basis pursuit denoise". Both approaches involve the solution of the basis pursuit problem that seeks a minimum one-norm solution of an underdetermined least-squares problem. Basis pursuit denoise (BPDN) fits the least-squares problem only approximately, and a single parameter determines a curve that traces the trade-off between the least-squares fit and the one-norm of the solution. In the work of Friedlander, it is shown show that the function that describes this curve is convex and continuously differentiable over all points of interest. They describe an efficient procedure for evaluating this function and its derivatives. As a result, they can compute arbitrary points on this curve. Their method is suitable for large-scale problems. Only matrix-vector operations are required. This is joint work with Ewout van der Berg and Michael P. Friedlander