On the Optimal Time Complexities in Decentralized Stochastic Asynchronous Optimization

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.

Authors

• Alexander Tyurin AIRI, Moscow, Russia Skoltech, Moscow, Russia
• Peter Richtárik King Abdullah University of Science and Technology

Keywords

No keywords are indexed for this paper.