Smoothing and worst-case complexity for direct-search methods in nonsmooth optimization
 
 
Description:  For smooth objective functions it has been shown that the worst-case cost of direct-search methods is of the same order as the one of steepest descent. Motivated by the lack of such a result in the nonsmooth case, we propose, analyze, and test a class of smoothing direct-search methods for the optimization of nonsmooth functions. Given a parameterized family of smoothing functions for the nonsmooth objective function, this class of methods consists of applying a direct search for a fixed value of the smoothing parameter until the step size is relatively small, after which the smoothing parameter is reduced and the process is repeated. One can show that the worst-case complexity (or cost) of this procedure is roughly one order of magnitude worse than the one for direct search or steepest descent on smooth functions. The class of smoothing direct-search methods is also showed to enjoy asymptotic global convergence properties. Numerical experience indicates that this approach leads to better values of the objective function, apparently without an additional cost in the number of function evaluations.
Date:  2012-10-10
Start Time:   11:30
Speaker:  Nima Garmanjani (Univ. Coimbra)
Institution:  FCT Scholarship SFRH/BD/33367/2008
Place:  Room 5.5 (DMUC)
Research Groups: -Numerical Analysis and Optimization
See more:   <Main>   <Lab for Computational Mathematics>  
 
© Centre for Mathematics, University of Coimbra, funded by
Science and Technology Foundation
Powered by: rdOnWeb v1.4 | technical support