By Dr.Luca Calatroni, investigador del laboratorio I3S de Sophia-Antipolis, Francia
Fecha seminario: 2021-11-11
We consider an inexact, scaled and adaptive Fast Iterative Soft-Thresholding Algorithm (FISTA) for minimising the sum of two (possibly strongly) convex functions. Inexactness is here explicitly taken into account to describe situations where proximal operators cannot be evaluated in closed form, while the idea of considering data-dependent scaling has been shown to be effective in incorporating Newton-type information along the iterations via suitable variable-metric updates. Finally, adaptivity is enforced by means of a non-monotone backtracking strategy improving the convergence speed compared to standard Armijoo-type approaches.