Closing the convergence gap of SGD without replacement

02/24/2020 ∙ by Shashank Rajput, et al. ∙ 0

Stochastic gradient descent without replacement sampling is widely used in practice for model training. However, the vast majority of SGD analyses assumes data sampled with replacement, and when the function minimized is strongly convex, an O(1/T) rate can be established when SGD is run for T iterations. A recent line of breakthrough work on SGD without replacement (SGDo) established an O(n/T^2) convergence rate when the function minimized is strongly convex and is a sum of n smooth functions, and an O(1/T^2+n^3/T^3) rate for sums of quadratics. On the other hand, the tightest known lower bound postulates an Ω(1/T^2+n^2/T^3) rate, leaving open the possibility of better SGDo convergence rates in the general case. In this paper, we close this gap and show that SGD without replacement achieves a rate of O(1/T^2+n^2/T^3) when the sum of the functions is a quadratic, and offer a new lower bound of Ω(n/T^2) for strongly convex functions that are sums of smooth functions.



There are no comments yet.


page 12

page 22

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.