Mathematics > Optimization and Control
[Submitted on 24 Mar 2018 (v1), revised 7 Jun 2018 (this version, v4), latest version 27 Nov 2018 (v5)]
Title:FROST -- Fast row-stochastic optimization with uncoordinated step-sizes
View PDFAbstract:In this paper, we discuss distributed optimization over directed graphs where doubly-stochastic weight matrices cannot be constructed. Most of the existing algorithms overcome this issue by applying push-sum consensus, which utilizes column-stochastic weights. The distributed implementation of column-stochastic weights requires each agent to know (at least) its out degree, which may be infeasible. In contrast, we describe FROST (Fast Row-stochastic Optimization with uncoordinated STep-sizes), an optimization algorithm applicable to directed graphs with row-stochastic weights and uncoordinated (non-identical) step-sizes among the agents. Its implementation is straightforward as each agent locally decides the weights assigned to the incoming information and locally chooses a suitable step-size. We show that FROST converges linearly to the global optimal when the largest step-size among the agents is sufficiently small and the objective functions are strongly-convex with Lipschitz-continuous gradients.
Submission history
From: Usman Khan [view email][v1] Sat, 24 Mar 2018 22:29:07 UTC (977 KB)
[v2] Sat, 31 Mar 2018 10:57:38 UTC (849 KB)
[v3] Tue, 17 Apr 2018 21:31:23 UTC (1,097 KB)
[v4] Thu, 7 Jun 2018 14:41:36 UTC (702 KB)
[v5] Tue, 27 Nov 2018 19:30:36 UTC (1,045 KB)
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.