Regularized Nonlinear Acceleration
Abstract
We describe a convergence acceleration technique for generic optimization problems. Our scheme computes estimates of the optimum from a nonlinear average of the iterates produced by any optimization method. The weights in this average are computed via a simple and small linear system, whose solution can be updated online. This acceleration scheme runs in parallel to the base algorithm, providing improved estimates of the solution on the fly, while the original optimization method is running. Numerical experiments are detailed on classical classification problems.
Cite
Text
Scieur et al. "Regularized Nonlinear Acceleration." Neural Information Processing Systems, 2016.Markdown
[Scieur et al. "Regularized Nonlinear Acceleration." Neural Information Processing Systems, 2016.](https://mlanthology.org/neurips/2016/scieur2016neurips-regularized/)BibTeX
@inproceedings{scieur2016neurips-regularized,
title = {{Regularized Nonlinear Acceleration}},
author = {Scieur, Damien and d'Aspremont, Alexandre and Bach, Francis},
booktitle = {Neural Information Processing Systems},
year = {2016},
pages = {712-720},
url = {https://mlanthology.org/neurips/2016/scieur2016neurips-regularized/}
}