EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
277. Accelerating Randomized Adaptive Subspace Trust-Region Derivative-Free Algorithms
Invited abstract in session TC-32: Algorithms for machine learning and inverse problems: zeroth-order optimisation, stream Advances in large scale nonlinear optimization.
Tuesday, 12:30-14:00Room: 41 (building: 303A)
Authors (first author is the speaker)
| 1. | Stefan M. Wild
|
| Applied Mathematics and Computational Research Division, Lawrence Berkeley National Laboratory | |
| 2. | Kwassi Joseph Dzahini
|
| Mathematics and Computer Science Division, Argonne National Laboratory | |
| 3. | Xiaoqian Liu
|
| The University of Texas MD Anderson Cancer Center |
Abstract
This talk presents an algorithm for solving deterministic derivative-free optimization problems with thousands of decision variables. We employ a randomized dimension reduction technique based on Johnson-Lindenstrauss transforms and the trust-region framework of Cartis and Roberts. A key ingredient in our work is to extend the affine space within which the algorithm operates to include information learned in the course of the optimization. By adapting this space and judiciously using information gained in the course of the optimization, we obtain significant performance benefits on a class of large-scale test problems without sacrificing convergence guarantees.
Keywords
- Mathematical Programming
- Continuous Optimization
- Programming, Nonlinear
Status: accepted
Back to the list of papers