English

Towards accelerated rates for distributed optimization over time-varying networks

Optimization and Control 2022-11-08 v5

Abstract

We study the problem of decentralized optimization over time-varying networks with strongly convex smooth cost functions. In our approach, nodes run a multi-step gossip procedure after making each gradient update, thus ensuring approximate consensus at each iteration, while the outer loop is based on accelerated Nesterov scheme. The algorithm achieves precision ε>0\varepsilon > 0 in O(κgχlog2(1/ε))O(\sqrt{\kappa_g}\chi\log^2(1/\varepsilon)) communication steps and O(κglog(1/ε))O(\sqrt{\kappa_g}\log(1/\varepsilon)) gradient computations at each node, where κg\kappa_g is the global function number and χ\chi characterizes connectivity of the communication network. In the case of a static network, χ=1/γ\chi = 1/\gamma where γ\gamma denotes the normalized spectral gap of communication matrix W\mathbf{W}. The complexity bound includes κg\kappa_g, which can be significantly better than the worst-case condition number among the nodes.

Keywords

Cite

@article{arxiv.2009.11069,
  title  = {Towards accelerated rates for distributed optimization over time-varying networks},
  author = {Alexander Rogozin and Vladislav Lukoshkin and Alexander Gasnikov and Dmitry Kovalev and Egor Shulgin},
  journal= {arXiv preprint arXiv:2009.11069},
  year   = {2022}
}
R2 v1 2026-06-23T18:44:28.254Z