![]() |
Universität Augsburg
|
![]() |
Herr Robert Mohr
Karlsruher Institut für Technologie
spricht am
Montag, 4. Dezember 2017
um
12:30 Uhr
im
Raum 3008 (L1)
über das Thema:
Abstract: |
We propose a new algorithm for solving large-scale non-convex machine learning problems. The algorithm is based on the classical trust-region framework and incorporates gradient and hessian information of the objective function obtained through subsampling. The subsample size is adaptively increased during the optimization process. We present a theoretical convergence result and compare the empirical performance of the algorithm on convex and non-convex machine learning problems to existing methods. |
Hierzu ergeht herzliche Einladung. |
Prof. Dr. Ralf Werner |