English

Comunication-Efficient Algorithms for Statistical Optimization

Machine Learning 2013-10-14 v3 Machine Learning Computation

Abstract

We analyze two communication-efficient algorithms for distributed statistical optimization on large-scale data sets. The first algorithm is a standard averaging method that distributes the NN data samples evenly to \nummac\nummac machines, performs separate minimization on each subset, and then averages the estimates. We provide a sharp analysis of this average mixture algorithm, showing that under a reasonable set of conditions, the combined parameter achieves mean-squared error that decays as \order(N1+(N/m)2)\order(N^{-1}+(N/m)^{-2}). Whenever mNm \le \sqrt{N}, this guarantee matches the best possible rate achievable by a centralized algorithm having access to all \totalnumobs\totalnumobs samples. The second algorithm is a novel method, based on an appropriate form of bootstrap subsampling. Requiring only a single round of communication, it has mean-squared error that decays as \order(N1+(N/m)3)\order(N^{-1} + (N/m)^{-3}), and so is more robust to the amount of parallelization. In addition, we show that a stochastic gradient-based method attains mean-squared error decaying as O(N1+(N/m)3/2)O(N^{-1} + (N/ m)^{-3/2}), easing computation at the expense of penalties in the rate of convergence. We also provide experimental evaluation of our methods, investigating their performance both on simulated data and on a large-scale regression problem from the internet search domain. In particular, we show that our methods can be used to efficiently solve an advertisement prediction problem from the Chinese SoSo Search Engine, which involves logistic regression with N2.4×108N \approx 2.4 \times 10^8 samples and d740,000d \approx 740,000 covariates.

Keywords

Cite

@article{arxiv.1209.4129,
  title  = {Comunication-Efficient Algorithms for Statistical Optimization},
  author = {Yuchen Zhang and John C. Duchi and Martin Wainwright},
  journal= {arXiv preprint arXiv:1209.4129},
  year   = {2013}
}

Comments

44 pages, to appear in Journal of Machine Learning Research (JMLR)

R2 v1 2026-06-21T22:07:39.068Z