Speaker: Sofia Zaourar, INRIA Grenoble, France
Location: ESB 4133
Intended Audience: Public
We consider convex nonsmooth optimization problems whose objective
function is known through some expensive procedure. For example, this
is the case in several problems that arise in electricity production
management, where the objective function is itself the result of an
optimization subproblem.
In this context, it often exists extra information – cheap but with
unknown accuracy – that is not used by the algorithms. In this talk,
we present a way to incorporate this coarse information into two
classical nonsmooth optimization algorithms: Kelley method and level
bundle method. We prove that the resulting methods are convergent and
we present numerical illustrations showing that they speed up
resolution.