GP warm-starting of dual decomposition – learning how to speed Up

Dual decomposition is a well-established algorithm to solve strongly convex optimization problems in decentralized fashion. However, as dual decomposition is a first-order method, it often takes a significant number of iterations to converge to acceptable accuracy. We propose a novel dual warm-starting strategy based on learning a predictor of the optimal dual variable of the consensus constraints. We use Gaussian process regression to learn the predictor from pre-computed data. Furthermore, we exploit the posterior variance error bounds of Gaussian Processes and draw upon numerical examples to illustrate the advantages of our approach.

Subjects

Distributed optimization

warm-starting

gaussian processes

machine learning

dual decomposition

posterior covariance

error bound

DDC Class

519: Applied Mathematics, Probabilities

006.31: Machine Learning

Lizenz

https://creativecommons.org/licenses/by-nc-nd/4.0/

Publication version

publishedVersion

Name

1-s2.0-S2405896326001345-main.pdf

Size

1014.91 KB

Format

Adobe PDF

Options

GP warm-starting of dual decomposition – learning how to speed Up