R-SPIDER: A Fast Riemannian Stochastic Optimization Algorithm with Curvature Independent Rate
Zhang, Jingzhao ; Zhang, Hongyi ; Sra, Suvrit
arXiv, 1811.04194 / Harvested from arXiv
Text on arXiv
We study smooth stochastic optimization problems on Riemannian manifolds. Via adapting the recently proposed SPIDER algorithm \citep{fang2018spider} (a variance reduced stochastic method) to Riemannian manifold, we can achieve faster rate than known algorithms in both the finite sum and stochastic settings. Unlike previous works, by \emph{not} resorting to bounding iterate distances, our analysis yields curvature independent convergence rates for both the nonconvex and strongly convex cases.
Publié le : 2018-11-09
Classification:  Mathematics - Optimization and Control,  Computer Science - Machine Learning 
@article{1811.04194,
     author = {Zhang, Jingzhao and Zhang, Hongyi and Sra, Suvrit},
     title = {R-SPIDER: A Fast Riemannian Stochastic Optimization Algorithm with
  Curvature Independent Rate},
     journal = {arXiv},
     volume = {2018},
     number = {0},
     year = {2018},
     language = {en},
     url = {http://dml.mathdoc.fr/item/1811.04194}
}

Zhang, Jingzhao; Zhang, Hongyi; Sra, Suvrit. R-SPIDER: A Fast Riemannian Stochastic Optimization Algorithm with
  Curvature Independent Rate. arXiv, Tome 2018 (2018) no. 0, . http://gdmltest.u-ga.fr/item/1811.04194/