Welcome to the homepage of the

Mathematical Optimization Group

Department of Mathematics, University of Tübingen, Germany

AFSI: Adaptive restart for fast semi-iterative schemes for convex optimisation

J.A. Tomasson, P. Ochs and J. Weickert

Abstract:
Smooth optimisation problems arise in many fields including image processing, and having fast methods for solving them has clear benefits. Widely and successfully used strategies to solve them are accelerated gradient methods. They accelerate standard gradient-based schemes by means of extrapolation. Unfortunately, most acceleration strategies are generic, in the sense, that they ignore specific information about the objective function. In this paper, we implement an adaptive restarting into a recently proposed efficient acceleration strategy that was coined Fast Semi-Iterative (FSI) scheme. Our analysis shows clear advantages of the adaptive restarting in terms of a theoretical convergence rate guarantee and state-of-the-art performance on a challenging image processing task.
pdf Bibtex
Citation:
J.A. Tomasson, P. Ochs, J. Weickert:
AFSI: Adaptive restart for fast semi-iterative schemes for convex optimisation. [pdf]
In T. Brox, A. Bruhn, M. Fritz (Eds.): German Conference on Pattern Recognition (GCPR). Lecture Notes in Computer Science, Vol. 11269, 669-681, Springer, 2019.
Bibtex:
@inproceedings{TOW19,
  title        = {AFSI: Adaptive restart for fast semi-iterative schemes for convex optimisation},
  author       = {J.A. Tomasson and P. Ochs and J. Weickert},
  year         = {2019},
  editor       = {T. Brox and A. Bruhn and M. Fritz},
  booktitle    = {German Conference on Pattern Recognition (GCPR)},
  series       = {Lecture Notes in Computer Science},
  publisher    = {Springer},
  volume       = {11269},
  pages        = {669--681}
}


MOP Group
©2017-2022
The author is not
responsible for
the content of
external pages.