Michael Garstka, Mark Cannon and Paul Goulart
in 2022 European Control Conference (ECC), London, UK, pp. 435-440, July 2022.
@inproceedings{GCG:2022,
author = {Garstka, Michael and Cannon, Mark and Goulart, Paul},
title = {Safeguarded Anderson acceleration for parametric nonexpansive operators},
booktitle = {2022 European Control Conference (ECC)},
year = {2022},
pages = {435-440},
url = {https://ieeexplore.ieee.org/abstract/document/9838320},
doi = {10.23919/ECC55457.2022.9838320}
}
This paper describes the design of a safeguarding scheme for Anderson acceleration to improve its practical performance and stability when used for first-order optimisation methods. We show how the combination of a non-expansiveness condition, conditioning constraints, and memory restarts integrate well with solver algorithms that can be represented as fixed point operators with dynamically varying parameters. The performance of the scheme is demonstrated on seven different QP and SDP problem types, including more than 500 problems. The safeguarded Anderson acceleration scheme proposed in this paper is implemented in the open-source ADMM-based conic solver COSMO.