Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enable the Right to be Forgotten with Federated Client Unlearning in Medical Imaging

Zhipeng Deng, Luyang Luo, Hao Chen

TL;DR

This paper addresses the challenge of implementing the right to be forgotten in federated learning for medical imaging. It introduces Federated Client Unlearning (FCU), combining Model-Contrastive Unlearning (MCU) for feature-level forgetting with Frequency-Guided Memory Preservation (FGMP) to retain global knowledge and enable rapid post-unlearning. FCU demonstrates superior forgetting efficacy and fidelity on intracranial hemorrhage and skin lesion tasks, achieving roughly a 10–15x speed-up over retraining from scratch. The proposed framework offers a practical path toward privacy-preserving, efficiently updatable FL models in medical imaging with minimal impact on generalization.

Abstract

The right to be forgotten, as stated in most data regulations, poses an underexplored challenge in federated learning (FL), leading to the development of federated unlearning (FU). However, current FU approaches often face trade-offs between efficiency, model performance, forgetting efficacy, and privacy preservation. In this paper, we delve into the paradigm of Federated Client Unlearning (FCU) to guarantee a client the right to erase the contribution or the influence, introducing the first FU framework in medical imaging. In the unlearning process of a client, the proposed model-contrastive unlearning marks a pioneering step towards feature-level unlearning, and frequency-guided memory preservation ensures smooth forgetting of local knowledge while maintaining the generalizability of the trained global model, thus avoiding performance compromises and guaranteeing rapid post-training. We evaluated our FCU framework on two public medical image datasets, including Intracranial hemorrhage diagnosis and skin lesion diagnosis, demonstrating that our framework outperformed other state-of-the-art FU frameworks, with an expected speed-up of 10-15 times compared with retraining from scratch. The code and the organized datasets can be found at: https://github.com/dzp2095/FCU.

Enable the Right to be Forgotten with Federated Client Unlearning in Medical Imaging

TL;DR

This paper addresses the challenge of implementing the right to be forgotten in federated learning for medical imaging. It introduces Federated Client Unlearning (FCU), combining Model-Contrastive Unlearning (MCU) for feature-level forgetting with Frequency-Guided Memory Preservation (FGMP) to retain global knowledge and enable rapid post-unlearning. FCU demonstrates superior forgetting efficacy and fidelity on intracranial hemorrhage and skin lesion tasks, achieving roughly a 10–15x speed-up over retraining from scratch. The proposed framework offers a practical path toward privacy-preserving, efficiently updatable FL models in medical imaging with minimal impact on generalization.

Abstract

The right to be forgotten, as stated in most data regulations, poses an underexplored challenge in federated learning (FL), leading to the development of federated unlearning (FU). However, current FU approaches often face trade-offs between efficiency, model performance, forgetting efficacy, and privacy preservation. In this paper, we delve into the paradigm of Federated Client Unlearning (FCU) to guarantee a client the right to erase the contribution or the influence, introducing the first FU framework in medical imaging. In the unlearning process of a client, the proposed model-contrastive unlearning marks a pioneering step towards feature-level unlearning, and frequency-guided memory preservation ensures smooth forgetting of local knowledge while maintaining the generalizability of the trained global model, thus avoiding performance compromises and guaranteeing rapid post-training. We evaluated our FCU framework on two public medical image datasets, including Intracranial hemorrhage diagnosis and skin lesion diagnosis, demonstrating that our framework outperformed other state-of-the-art FU frameworks, with an expected speed-up of 10-15 times compared with retraining from scratch. The code and the organized datasets can be found at: https://github.com/dzp2095/FCU.
Paper Structure (12 sections, 2 equations, 2 figures, 2 tables)

This paper contains 12 sections, 2 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Preliminaries and Overview
  4. Model-Contrastive Unlearning
  5. Frequency-Guided Memory Preservation
  6. Overall of FCU framework
  7. Experiments
  8. Experiment Setup
  9. Comparison with state-of-the-arts
  10. Conclusion
  11. Acknowledgments.
  12. Disclosure of Interests.

Figures (2)

  • Figure 1: Overview of Federated Client Unlearning (FCU): The target client conducts unlearning locally and updates the server with the unlearned model, which then serves as the initial model for post-training on the remaining clients.
  • Figure 2: Ablation study of FGMP and MCU across local unlearning iterations, showing forgotten $Error^f$ and Accuracy for the two tasks.