Mathematics > Optimization and Control
[Submitted on 17 Mar 2024]
Title:Accelerating Gradient Tracking with Periodic Global Averaging
View PDF HTML (experimental)Abstract:Decentralized optimization algorithms have recently attracted increasing attention due to its wide applications in all areas of science and engineering. In these algorithms, a collection of agents collaborate to minimize the average of a set of heterogeneous cost functions in a decentralized manner. State-of-the-art decentralized algorithms like Gradient Tracking (GT) and Exact Diffusion (ED) involve communication at each iteration. Yet, communication between agents is often expensive, resource intensive, and can be very slow. To this end, several strategies have been developed to balance between communication overhead and convergence rate of decentralized methods. In this paper, we introduce GT-PGA, which incorporates~GT with periodic global averaging. With the additional PGA, the influence of poor network connectivity in the GT algorithm can be compensated or controlled by a careful selection of the global averaging period. Under the stochastic, nonconvex setup, our analysis quantifies the crucial trade-off between the connectivity of network topology and the PGA period. Thus, with a suitable design of the PGA period, GT-PGA improves the convergence rate of vanilla GT. Numerical experiments are conducted to support our theory, and simulation results reveal that the proposed GT-PGA accelerates practical convergence, especially when the network is sparse.
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.