Martin Zinkevich, Markus Weimer, Alex Smola, Lihong Li

Abstract

With the increase in available data parallel machine learning has become an increasingly pressing problem. In this paper we present the first parallel stochastic gradient descent algorithm including a detailed analysis and experimental evidence. Unlike prior work on parallel optimization algorithms [5, 7] our variant comes with parallel acceleration guarantees and it poses no overly tight latency constraints, which might only be available in the multicore setting. Our analysis introduces a novel proof technique — contractive mappings to quantify the speed of convergence of parameter distributions to their asymptotic limits. As a side effect this answers the question of how quickly stochastic gradient descent algorithms reach the asymptotically normal regime [1, 8].

Download PDF of Paper and Appendix

BibTeX

@inproceedings{NIPS2010_1162, 
  author = {Martin Zinkevich and Markus Weimer and Alex Smola and Lihong Li}, 
  Booktitle = {Advances in Neural Information Processing Systems 23}, 
  Editor = {J. Lafferty and C. K. I. Williams and J. Shawe-Taylor and R.S. Zemel and A. Culotta}, 
  Pages = {2595--2603}, 
  Title = {Parallelized Stochastic Gradient Descent}, 
  Year = {2010}
}