942. Stepsizes for the gradient method based on subspace approximations
Invited abstract in session MC-37: Optimization algorithms and applications 1, stream Nonlinear Optimization.
Monday, 10:30-12:00Room: JUR – Seminar-Raum 43
Authors (first author is the speaker)
| 1. | Ya-xiang Yuan
|
| Institute of Computational Mathematics and Sci./Eng. Computing, AMSS, Chinese Academy of Sciences |
Abstract
The gradient method is the simpliest method for unconstrained optimization, as in each iteration, the next iteration point is obtained from the current one by moving along the gradient direction. Variations of the gradient methods, such as stochastic gradient algorithms are widely used for machine learning and other AI problems. In this talk, we give some new stepsizes
for the gradient method based on subspace approximations, which presents a new approach for constructing gradient algorithms.
Keywords
- Programming, Nonlinear
- Optimization Models and Methods
- Machine Learning
Status: accepted
Back to the list of papers