Communication-Efficient Zero-Order and First-Order Federated Learning Methods over Wireless Networks

Mohamad Assaad, Senior, IEEE, Zeinab Nehme, Student, IEEE, Merouane Debbah, Fellow, IEEE · 2025

Read Communication-Efficient Zero-Order and First-Order Federated Learning Methods over Wireless Networks by Mohamad Assaad, Senior, IEEE, Zeinab Nehme, Student, IEEE, Merouane Debbah, Fellow, IEEE (2025) 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-2508-08013.txt.

Description

Mohamad Assaad, Zeinab Nehme, and Merouane Debbah design two communication-light federated learning methods for wireless networks, in a 2025 paper. Federated learning trains models across edge devices without moving their data, but the training traffic can swamp a wireless link. Their methods send single numbers instead of long vectors, and let many users transmit at once.

One method estimates gradients from two probes, with no gradient math at the device. The other computes gradients directly. Both fold channel knowledge into the learning itself, so no extra resources go to measuring the channel, and devices need not stay in sync. The authors prove convergence for both and bound their performance.

Catalog record

CREATORMohamad Assaad, Senior, IEEE, Zeinab Nehme, Student, IEEE, Merouane Debbah, Fellow, IEEE
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE79 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-08013

Browse the full library · About the Tower of Babel