A Unified and Refined Convergence Analysis for Non-Convex Decentralized Learning

TL;DR

This paper provides a unified convergence analysis for decentralized non-convex optimization algorithms, showing they are less sensitive to network topology than previously thought, aligning theory with empirical observations.

Contribution

The authors introduce a general stochastic unified decentralized algorithm (SUDA) and establish improved convergence bounds that better reflect the algorithms' robustness to network topology.

Findings

SUDA converges under non-convex and Polyak-Lojasiewicz conditions.

Decentralized methods like Exact-Diffusion are less sensitive to network topology than DSGD.

Results align with empirical experiments showing robustness of certain algorithms.

Abstract

We study the consensus decentralized optimization problem where the objective function is the average of $n$ agents private non-convex cost functions; moreover, the agents can only communicate to their neighbors on a given network topology. The stochastic learning setting is considered in this paper where each agent can only access a noisy estimate of its gradient. Many decentralized methods can solve such problem including EXTRA, Exact-Diffusion/D$^2$, and gradient-tracking. Unlike the famed DSGD algorithm, these methods have been shown to be robust to the heterogeneity across the local cost functions. However, the established convergence rates for these methods indicate that their sensitivity to the network topology is worse than DSGD. Such theoretical results imply that these methods can perform much worse than DSGD over sparse networks, which, however, contradicts empirical experiments where DSGD is observed to be more sensitive to the network topology. In this work, we study a general stochastic unified decentralized algorithm (SUDA) that includes the above methods as special cases. We establish the convergence of SUDA under both non-convex and the Polyak-Lojasiewicz condition settings. Our results provide improved network topology dependent bounds for these methods (such as Exact-Diffusion/D$^2$ and gradient-tracking) compared with existing literature. Moreover, our results show that these methods are often less sensitive to the network topology compared to DSGD, which agrees with numerical experiments.