FedUP: One-Shot Federated Unlearning via Centroid-Guided Plug-in Filters

2026-06-22

AI Summary

This work proposes a single-round federated unlearning framework to address the challenges of unintended knowledge loss and high latency caused by user data deletion requests in federated learning, thereby complying with regulatory requirements such as the “right to be forgotten.” The approach trains lightweight, pluggable filters at the server while keeping the original model parameters frozen. These filters are guided by differentially private class centroid samples to precisely suppress representations associated with the target data. Notably, the architecture eliminates the need for multi-round communication or full model retraining and offers inherent reversibility, enabling unlearning operations to complete within seconds. Experimental results across diverse image and text tasks demonstrate that the method significantly reduces collateral knowledge loss while achieving high accuracy and efficiency, reducing unlearning latency from minutes to seconds.

Abstract

Federated unlearning (FU) is critical for complying with legal mandates like the right to be forgotten in decentralized systems, yet current methods face a persistent dilemma between non-target knowledge loss and high request latency. To resolve these issues, we propose FedUP, a one-shot federated unlearning framework utilizing lightweight pluggable filters that act as a "knowledge funnel" to screen out target data while preserving original model performance. By freezing original model parameters and training filters at the server side using differentially private (DP)-protected class centroid samples, FedUP bypasses the need for multi-round client-server communication and complex retraining, reducing unlearning latency from minutes to mere seconds. Additionally, the framework's pluggable architecture ensures inherent reversibility, enabling the seamless restoration of forgotten knowledge by simply removing the filters. Extensive experiments on diverse image and text tasks demonstrate that FedUP effectively reduces non-target knowledge loss and achieves superior unlearning precision and efficiency across various scenarios. Code is available at: https://github.com/suows/FedUP-code.

Problem

• federated unlearning
• right to be forgotten
• knowledge loss
• request latency
• decentralized systems

Innovation

• federated unlearning
• one-shot unlearning
• pluggable filters
• centroid-guided
• differentially private