Resource-Adaptive Federated Text Generation with Differential Privacy

Jiayi Wang, John Gounley, Heidi Hanson · 2026

Read Resource-Adaptive Federated Text Generation with Differential Privacy by Jiayi Wang, John Gounley, Heidi Hanson (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-2603-07027.txt.

Description

Wang and colleagues propose resource-adaptive federated text generation with differential privacy. In cross-silo learning, sensitive text stays local under privacy rules, so retraining per task costs much bandwidth and privacy budget. A fix is private synthetic data that mimics the global mix for reuse across tasks. Yet pretrained models fail under domain shift, and uneven compute blocks weak clients from tuning, worsening skew and noise. Their framework adapts to client power: strong clients tune privately, while weak clients vote lightly to refine synthetic data.

Catalog record

CREATORJiayi Wang, John Gounley, Heidi Hanson
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE63963
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.07027

Browse the full library · About the Tower of Babel