Alert button

Decentralized Gradient Tracking with Local Steps

Jan 03, 2023
Yue Liu, Tao Lin, Anastasia Koloskova, Sebastian U. Stich

Figure 1 for Decentralized Gradient Tracking with Local Steps
Figure 2 for Decentralized Gradient Tracking with Local Steps
Figure 3 for Decentralized Gradient Tracking with Local Steps
Figure 4 for Decentralized Gradient Tracking with Local Steps

Share this with someone who'll enjoy it:

Gradient tracking (GT) is an algorithm designed for solving decentralized optimization problems over a network (such as training a machine learning model). A key feature of GT is a tracking mechanism that allows to overcome data heterogeneity between nodes. We develop a novel decentralized tracking mechanism, $K$-GT, that enables communication-efficient local updates in GT while inheriting the data-independence property of GT. We prove a convergence rate for $K$-GT on smooth non-convex functions and prove that it reduces the communication overhead asymptotically by a linear factor $K$, where $K$ denotes the number of local steps. We illustrate the robustness and effectiveness of this heterogeneity correction on convex and non-convex benchmark problems and on a non-convex neural network training task with the MNIST dataset.

View paper onarxiv icon

Share this with someone who'll enjoy it: