English

On the Optimality of Averaging in Distributed Statistical Learning

Machine Learning 2016-06-14 v2 Statistics Theory Statistics Theory

Abstract

A common approach to statistical learning with big-data is to randomly split it among mm machines and learn the parameter of interest by averaging the mm individual estimates. In this paper, focusing on empirical risk minimization, or equivalently M-estimation, we study the statistical error incurred by this strategy. We consider two large-sample settings: First, a classical setting where the number of parameters pp is fixed, and the number of samples per machine nn\to\infty. Second, a high-dimensional regime where both p,np,n\to\infty with p/nκ(0,1)p/n \to \kappa \in (0,1). For both regimes and under suitable assumptions, we present asymptotically exact expressions for this estimation error. In the fixed-pp setting, under suitable assumptions, we prove that to leading order averaging is as accurate as the centralized solution. We also derive the second order error terms, and show that these can be non-negligible, notably for non-linear models. The high-dimensional setting, in contrast, exhibits a qualitatively different behavior: data splitting incurs a first-order accuracy loss, which to leading order increases linearly with the number of machines. The dependence of our error approximations on the number of machines traces an interesting accuracy-complexity tradeoff, allowing the practitioner an informed choice on the number of machines to deploy. Finally, we confirm our theoretical analysis with several simulations.

Keywords

Cite

@article{arxiv.1407.2724,
  title  = {On the Optimality of Averaging in Distributed Statistical Learning},
  author = {Jonathan Rosenblatt and Boaz Nadler},
  journal= {arXiv preprint arXiv:1407.2724},
  year   = {2016}
}

Comments

Major changes from previous version. Particularly on the second order error approximation and implications

R2 v1 2026-06-22T05:00:22.806Z