We describe ASAGA, an asynchronous parallel version of the incremental gradient algorithm SAGA that enjoys fast linear convergence rates. Through a novel perspective, we revisit and clarify a subtle but important technical issue present in a large fraction of the recent convergence rate proofs for asynchronous parallel optimization algorithms, and propose a simplification of the recently introduced "perturbed iterate" framework that resolves it. We thereby prove that ASAGA can obtain a theoretical linear speedup on multi-core systems even without sparsity assumptions. We present results of an implementation on a 40-core architecture illustrating the practical speedup as well as the hardware overhead.
@article{arxiv.1606.04809,
title = {ASAGA: Asynchronous Parallel SAGA},
author = {Rémi Leblond and Fabian Pedregosa and Simon Lacoste-Julien},
journal= {arXiv preprint arXiv:1606.04809},
year = {2017}
}
Comments
Appears in: Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS 2017), 37 pages