Distributed Stochastic Gradient Tracking Methods

Abstract

In this paper, we study the problem of distributed multi-agent optimization over a network, where each agent possesses a local cost function that is smooth and strongly convex. The global objective is to find a common solution that minimizes the average of all cost functions. Assuming agents only have access to unbiased estimates of the gradients of their local cost functions, we consider a distributed stochastic gradient tracking method (DSGT) and a gossip-like stochastic gradient tracking method (GSGT). We show that, in expectation, the iterates generated by each agent are attracted to a neighborhood of the optimal solution, where they accumulate exponentially fast (under a constant stepsize choice). Under DSGT, the limiting (expected) error bounds on the distance of the iterates from the optimal solution decrease with the network size $n$, which is a comparable performance to a centralized stochastic gradient algorithm. Moreover, we show that when the network is well-connected, GSGT incurs lower communication cost than DSGT while maintaining a similar computational cost. Numerical example further demonstrates the effectiveness of the proposed methods.

Accepted in Mathematical Programming. This article draws heavily from arXiv:1803.07741 (conference submission)

References in corpus (2)

• A Sharp Estimate on the Transient Time of Distributed Stochastic Gradient Descent
• A Push-Pull Gradient Method for Distributed Optimization in Networks

Cited by in corpus (6)

• Multi-Agent Reinforcement Learning via Double Averaging Primal-Dual Optimization
• A Push-Pull Gradient Method for Distributed Optimization in Networks
• Distributed Deep Learning with Event-Triggered Communication
• Asymptotic Network Independence in Distributed Stochastic Optimization for Machine Learning
• A Distributed Stochastic Gradient Tracking Method
• Nested Distributed Gradient Methods with Stochastic Computation Errors