Skip to yearly menu bar Skip to main content


Poster

On the Optimal Time Complexities in Decentralized Stochastic Asynchronous Optimization

Alexander Tyurin · Peter Richtarik

Poster Room - TBD
[ ]
Fri 13 Dec 4:30 p.m. PST — 7:30 p.m. PST

Abstract:

We consider the decentralized stochastic asynchronous optimization setup, where many workers asynchronously calculate stochastic gradients and asynchronously communicate with each other using edges in a multigraph. For both homogeneous and heterogeneous setups, we prove new time complexity lower bounds under the assumption that computation and communication speeds are bounded by constants. After that, we developed a new nearly optimal method, Fragile SGD, and a new optimal method, Amelie SGD, that converge with arbitrary heterogeneous computation and communication speeds and match our lower bounds (up to a logarithmic factor in the homogeneous setting). Our time complexities are new, nearly optimal, and provably improve all previous asynchronous/synchronous stochastic methods in the decentralized setup.

Live content is unavailable. Log in and register to view live content