Accelerated Gradient Tracking over Time-varying Graphs for Decentralized Optimization
share
Decentralized optimization over time-varying graphs has been increasingly common in modern machine learning with massive data stored on millions of mobile devices, such as in federated learning. This paper revisits and extends the widely used accelerated gradient tracking. We prove the O(γ^2/(1-σ_γ)^2√(L/ϵ)) and O((γ/1-σ_γ)^1.5√(L/μ)log1/ϵ) complexities for the practical single loop accelerated gradient tracking over time-varying graphs when the problems are nonstrongly convex and strongly convex, respectively, where γ and σ_γ are two common constants charactering the network connectivity, ϵ is the desired precision, and L and μ are the smoothness and strong convexity constants, respectively. Our complexities improve significantly on the ones of O(1/ϵ^5/7) and O((L/μ)^5/71/(1-σ)^1.5log1/ϵ) proved in the original literature only for static graph. When combining with a multiple consensus subroutine, the dependence on the network connectivity constants can be further improved. When the network is time-invariant, our complexities exactly match the lower bounds without hiding any poly-logarithmic factor for both nonstrongly convex and strongly convex problems.
READ FULL TEXT