Personalized Federated Learning via Dual-Prompt Optimization and Cross Fusion

Yuguang Zhang, Kuangpu Guo, Zhihe Lu, Yunbo Wang, Jian Liang · 2025

Read Personalized Federated Learning via Dual-Prompt Optimization and Cross Fusion by Yuguang Zhang, Kuangpu Guo, Zhihe Lu, Yunbo Wang, Jian Liang (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-2506-21144.txt.

Description

Yuguang Zhang and co-authors personalize federated learning with a double set of prompts, one for words and one for images. Federated training lets many clients build one model without pooling data, but each client's data shifts in both labels and domain. Prompt tuning of vision-language models is cheap, yet earlier federated versions tuned text prompts only.

Their framework, pFedDC, keeps global and local prompts in both modalities on every client: global ones carry shared knowledge, local ones encode each client's quirks. A cross-fusion module then blends the levels adaptively for each client. Across nine data sets with varied heterogeneity, the method consistently beat prior state-of-the-art systems. The authors frame dual prompts as the missing half of federated personalization.

Catalog record

CREATORYuguang Zhang, Kuangpu Guo, Zhihe Lu, Yunbo Wang, Jian Liang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-21144

Browse the full library · About the Tower of Babel