E-3SFC: Communication-Efficient Federated Learning with Double-way Features Synthesizing

Yuhao Zhou, Yuxin Tian, Mingjia Shi, Yuanxi Li, Yanan Sun, Qing Ye, Jiancheng Lv · 2025

Read E-3SFC: Communication-Efficient Federated Learning with Double-way Features Synthesizing by Yuhao Zhou, Yuxin Tian, Mingjia Shi, Yuanxi Li, Yanan Sun, Qing Ye, Jiancheng Lv (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-2502-03092.txt.

Description

Yuhao Zhou and co-authors published this machine learning paper in 2025. It works on federated learning, where many devices train one shared model without sending their raw data to a central server. The catch is traffic: as models grow, shipping updates back and forth becomes a heavy burden. Squeezing those updates down usually adds errors that slow learning.

The team's method, called E-3SFC, treats gradient computation in reverse. It views the model itself as a device that can rebuild full gradients from tiny synthetic features, so only those features need to travel. The scheme compresses in both directions and adjusts the traffic budget as training proceeds. Across six datasets and six models, it beat leading methods by up to 13.4 percent while cutting communication costs by more than a hundredfold.

Catalog record

CREATORYuhao Zhou, Yuxin Tian, Mingjia Shi, Yuanxi Li, Yanan Sun, Qing Ye, Jiancheng Lv
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE89 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-03092

Browse the full library · About the Tower of Babel