Federated learning with prototype-based adaptive domain adjustment

doi:10.1016/j.neucom.2026.133967

Federated learning with prototype-based adaptive domain adjustment

2026-05-14

0

PRE

AI

DOI：10.1016/j.neucom.2026.133967

Original

Original request for help

Save

Abstract

En 中文

• We propose FedPAD, a prototype-based method for federated learning under domain shift. • The server learns dynamic weights for prototype aggregation to improve domain coverage. • A separate temperature-controlled mapping stabilizes model aggregation under client drift. • We provide a convergence bound with an explicit term for weight variation across rounds. • Experiments on four domain-shift benchmarks show improved average and worst-domain accuracy.

Keywords:

FedPAD

prototype-based method

domain shift

federated learning

adaptive domain adjustment

Federated learning with prototype-based adaptive domain adjustment

Abstract

Journal

Organization