ICML 2017poster52 citations

Stochastic Convex Optimization: Faster Local Growth Implies Faster Global Convergence

Yi Xu, Qihang Lin, Tianbao Yang

Abstract

In this paper, a new theory is developed for first-order stochastic convex optimization, showing that the global convergence rate is sufficiently quantified by a local growth rate of the objective function in a neighborhood of the optimal solutions. In particular, if the objective function $F(\mathbf{w})$ in the $\epsilon$-sublevel set grows as fast as $\|\mathbf{w} - \mathbf{w}_*\|_2^{1/\theta}$, where $\mathbf{w}_*$ represents the closest optimal solution to $\mathbf{w}$ and $\theta\in(0,1]$ quantifies the local growth rate, the iteration complexity of first-order stochastic optimization for achieving an $\epsilon$-optimal solution can be $\widetilde O(1/\epsilon^{2(1-\theta)})$, which is

BibTeX
@InProceedings{pmlr-v70-xu17a,
  title = 	 {Stochastic Convex Optimization: Faster Local Growth Implies Faster Global Convergence},
  author =       {Yi Xu and Qihang Lin and Tianbao Yang},
  booktitle = 	 {Proceedings of the 34th International Conference on Machine Learning},
  pages = 	 {3821--3830},
  year = 	 {2017},
  editor = 	 {Precup, Doina and Teh, Yee Whye},
  volume = 	 {70},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {06--11 Aug},
  publisher =    {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v70/xu17a/xu17a.pdf},
  url = 	 {https://proceedings.mlr.press/v70/xu17a.html},
  abstract = 	 {In this paper, a new theory is developed for first-order stochastic convex optimization, showing that the global convergence rate is sufficiently quantified by a local growth rate of the objective function in a neighborhood of the optimal solutions. In particular, if the objective function $F(\mathbf{w})$ in the $\epsilon$-sublevel set grows as fast as $\|\mathbf{w} - \mathbf{w}_*\|_2^{1/\theta}$, where $\mathbf{w}_*$ represents the closest optimal solution to $\mathbf{w}$ and $\theta\in(0,1]$ quantifies the local growth rate, the iteration complexity of first-order stochastic optimization for achieving an $\epsilon$-optimal solution can be $\widetilde O(1/\epsilon^{2(1-\theta)})$, which is
Stochastic Convex Optimization: Faster Local Growth Implies Faster Global Convergence · ICML 2017