Over the past decades, driven by the success of the Heavy-ball method, Nesterov's acceleration and FISTA scheme, inertial technique has become one of the most popular approaches for accelerating optimization methods. However, the performance of other inertial "accelerated" first-order methods, such as inertial Douglas--Rachford, is not as good as the aforementioned ones, or even results in deacceleration. In this talk, I will first discuss the limitations of inertial technique, particular its failures. Then I will present a geometry inspired acceleration scheme. Both relation with existing work and numerical experiments will be covered.
This is a joint work with Clarice Poon.
On the Acceleration of First-Order Methods and a New One
25.10.2021 15:30 - 16:30
Organiser:
R. I. Boț (U Wien), S. Sabach (Technion - Israel Institute of Technology Haifa), M. Staudigl (Maastricht U)
Location:
Zoom Meeting
Verwandte Dateien
- abstract_Liang_OWOS.pdf 146 KB