pith. sign in

arxiv: 1308.3558 · v1 · pith:FCHQYPYGnew · submitted 2013-08-16 · 💻 cs.LG · cs.NA· math.NA

Fast Stochastic Alternating Direction Method of Multipliers

classification 💻 cs.LG cs.NAmath.NA
keywords admmalgorithmstochasticalgorithmsalternatingbatchconvergencedirection
0
0 comments X
read the original abstract

In this paper, we propose a new stochastic alternating direction method of multipliers (ADMM) algorithm, which incrementally approximates the full gradient in the linearized ADMM formulation. Besides having a low per-iteration complexity as existing stochastic ADMM algorithms, the proposed algorithm improves the convergence rate on convex problems from $O(\frac 1 {\sqrt{T}})$ to $O(\frac 1 T)$, where $T$ is the number of iterations. This matches the convergence rate of the batch ADMM algorithm, but without the need to visit all the samples in each iteration. Experiments on the graph-guided fused lasso demonstrate that the new algorithm is significantly faster than state-of-the-art stochastic and batch ADMM algorithms.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.