FedCARE: Federated Unlearning with Conflict-Aware Projection and Relearning-Resistant Recovery

TL;DR

This paper tackles the challenge of enforcing privacy in federated learning under the right-to-be-forgotten by proposing FedCARE, a unified, low-overhead federated unlearning framework. FedCARE combines a data-free pseudo-sample generator, conflict-aware projection unlearning, and relearning-resistant recovery to remove forgotten data influence while preserving retained knowledge, even under non-IID data. The approach supports client-, instance-, and class-level unlearning and demonstrates strong forgetting performance, utility retention, and security against membership inference and backdoors across MNIST, SVHN, CIFAR-10, and CIFAR-100. Empirical results show FedCARE achieves effective forgetting with minimal overhead compared to retraining and outperforms state-of-the-art FU baselines, highlighting its practical potential for privacy-compliant FL systems.

Abstract

Federated learning (FL) enables collaborative model training without centralizing raw data, but privacy regulations such as the right to be forgotten require FL systems to remove the influence of previously used training data upon request. Retraining a federated model from scratch is prohibitively expensive, motivating federated unlearning (FU). However, existing FU methods suffer from high unlearning overhead, utility degradation caused by entangled knowledge, and unintended relearning during post-unlearning recovery. In this paper, we propose FedCARE, a unified and low overhead FU framework that enables conflict-aware unlearning and relearning-resistant recovery. FedCARE leverages gradient ascent for efficient forgetting when target data are locally available and employs data free model inversion to construct class level proxies of shared knowledge. Based on these insights, FedCARE integrates a pseudo-sample generator, conflict-aware projected gradient ascent for utility preserving unlearning, and a recovery strategy that suppresses rollback toward the pre-unlearning model. FedCARE supports client, instance, and class level unlearning with modest overhead. Extensive experiments on multiple datasets and model architectures under both IID and non-IID settings show that FedCARE achieves effective forgetting, improved utility retention, and reduced relearning risk compared to state of the art FU baselines.

Figures (4)

• Figure 1: Overall framework illustrated with client-level unlearning. After federated training, the server trains a lightweight data-free generator once. Upon an unlearning request, the generator is sent to the target client, which performs conflict-aware projection unlearning using its private forget set and pseudo-samples. The remaining clients then conduct constrained recovery with relearning-resistant aggregation.
• Figure 2: Client-level unlearning performance across different classes on CIFAR-10 (non-IID). The ($*$) denotes classes contained within the unlearning client.
• Figure 3: Class-level unlearning performance across different classes on CIFAR-10 (non-IID). The ($*$) denotes the class to be forgotten.
• Figure 4: Efficiency comparison between FedCARE and baseline methods.