Dynamic Topology Optimization for Non-IID Data in Decentralized Learning

Bart Cox, Antreas Ioannou, Jérémie Decouchant · 2026

Read Dynamic Topology Optimization for Non-IID Data in Decentralized Learning by Bart Cox, Antreas Ioannou, Jérémie Decouchant (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03383.txt.

Description

In decentralized learning, many machines train one model together without any central coordinator. Bart Cox and 2 co-authors address a weak point of that setup in a 2026 paper. Accuracy drops when each node holds different kinds of data and the communication network never changes. Their answer is an algorithm named Morph.

Morph lets each node pick exchange partners whose models differ most from its own, while keeping the number of incoming links fixed. Nodes find new peers through gossip and keep reshaping the network during training. Tests on image datasets with up to 100 nodes beat static and epidemic baselines. They come close to a fully connected network, converge faster, and need fewer communication rounds.

Catalog record

CREATORBart Cox, Antreas Ioannou, Jérémie Decouchant
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-03383

Browse the full library · About the Tower of Babel