Several optimisation algorithms, including ADMM and Douglas-Rachford algorithm (DR), can be cast as the fixed-point iteration (FPI) of an averaged operator. However, these algorithms' convergence is generally slow (within sublinear regime).
- Speaker
- Date
- Friday 18 Oct 2024, 12:00 - 13:00
- Type
- Seminar
- Room
- ET-14
- Building
- E Building
- More information
Lunch will be provided (vegetarian option included).
For more information please contact the Secretariat Econometrics at eb-secr@ese.eur.nl
We show that the linear convergence of the FPI of averaged operators is equivalent to an error-bound condition on the operator. Our work captures several existing results on the linear convergence of the ADMM and DR under weaker assumptions. As a byproduct of our method, we bound the rate of convergence of the DR applied to linear and quadratic optimisation problems.
About the speaker
Juan C. Vera Lizcano is a professor of optimisation in the department of Econometrics and Operations Research at Tilburg University. His expertise includes polynomial and conic optimisation and their applications to combinatorial optimization; relation between error bounds, condition numbers, and convergence of optimisation algorithms; sparse methods for machine learning.
