MICS seminar: Adam Ouorou
--Adam Ouorou (Orange) : Non-smooth convex optimization in practice.
Jueves 3 abril, 14:00Pasado
Non-smooth convex optimization in practice.
We review the main methods for the minimization problems of non-smooth convex functions. The simplest methods are based on constructing models of the function to be minimized. These models are not reliable even close to an optimum, because of the lack of continuity of first-order information. This motivate the introduction of different stabilization methods. The methods based on the so-called localization sets are presented as well as their improvements by stabilization. We conclude with recent approaches using acceleration techniques proposed by Nesterov for gradient methods.
