FedEMA-Distill: Exponential Moving Average Guided Knowledge Distillation for Robust Federated Learning

Hamza Reguieg, Mohamed El Kamili, Essaid Sabir · 2026

Read FedEMA-Distill: Exponential Moving Average Guided Knowledge Distillation for Robust Federated Learning by Hamza Reguieg, Mohamed El Kamili, Essaid Sabir (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-2603-04422.txt.

Description

Reguieg and colleagues harden federated learning with FedEMA-Distill. Mixed client data and adversarial clients cause drift, slow convergence, and heavy traffic. Clients train locally and upload only compressed logits, keeping their own architectures. The server blends a moving average global model with ensemble distillation on a small public set. Tests on CIFAR, FEMNIST, and AG News under strong skew lift accuracy by several points. Targets arrive in 30 to 35 percent fewer rounds, payloads stay under half a megabyte, and robust aggregation steadies training under attack.

Catalog record

CREATORHamza Reguieg, Mohamed El Kamili, Essaid Sabir
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE56768
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.04422

Browse the full library · About the Tower of Babel