FUPareto: Bridging the Forgetting-Utility Gap in Federated Unlearning via Pareto Augmented Optimization PubID: pubid: 0000–0000/00$00.00 © 2021 IEEE

Zeyan Wang, Zhengmao Liu, Yongxin Cai, Chi Li, Xiaoying Tang, Jingchao Chen, Zibin Pan*, Jing Qiu · 2026

Read FUPareto: Bridging the Forgetting-Utility Gap in Federated Unlearning via Pareto Augmented Optimization PubID: pubid: 0000–0000/00$00.00 © 2021 IEEE by Zeyan Wang, Zhengmao Liu, Yongxin Cai, Chi Li, Xiaoying Tang, Jingchao Chen, Zibin Pan*, Jing Qiu (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-2602-01852.txt.

Description

Zeyan Wang and colleagues remove client data from a federated model, tackling utility and leak risks. Multi client erasure adds gradient conflicts that degrade forgetting, so their FUPareto framework uses Pareto optimization. An MBS loss pushes the target logit under the best non target logit, improving efficiency and cutting inference risk. Pareto gains keep utility, Pareto expansion guarantees forgetting, and a null space projected gradient method decouples client gradient conflicts. Tests show it beats federated unlearning methods on both efficacy and retained utility.

Catalog record

CREATORZeyan Wang, Zhengmao Liu, Yongxin Cai, Chi Li, Xiaoying Tang, Jingchao Chen, Zibin Pan*, Jing Qiu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38344
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.01852

Browse the full library · About the Tower of Babel