English

Optimal Distributed Optimization on Slowly Time-Varying Graphs

Optimization and Control 2019-12-02 v7

Abstract

We study optimal distributed first-order optimization algorithms when the network (i.e., communication constraints between the agents) changes with time. This problem is motivated by scenarios where agents experience network malfunctions. We provide a sufficient condition that guarantees a convergence rate with optimal (up lo logarithmic terms) dependencies on the network and function parameters if the network changes are constrained to a small percentage α\alpha of the total number of iterations. We call such networks slowly time-varying networks. Moreover, we show that Nesterov's method has an iteration complexity of Ω((κΦχˉ+αlog(κΦχˉ))log(1/ε))\Omega \big( \big(\sqrt{\kappa_\Phi \cdot \bar{\chi}} + \alpha \log(\kappa_\Phi \cdot \bar{\chi})\big) \log(1 / \varepsilon)\big) for decentralized algorithms, where κΦ\kappa_\Phi is condition number of the objective function, and χˉ\bar\chi is a worst case bound on the condition number of the sequence of communication graphs. Additionally, we provide an explicit upper bound on α\alpha in terms of the condition number of the objective function and network topologies.

Keywords

Cite

@article{arxiv.1805.06045,
  title  = {Optimal Distributed Optimization on Slowly Time-Varying Graphs},
  author = {Alexander Rogozin and César A. Uribe and Alexander Gasnikov and Nikolay Malkovsky and Angelia Nedić},
  journal= {arXiv preprint arXiv:1805.06045},
  year   = {2019}
}
R2 v1 2026-06-23T01:56:46.458Z