A Revocation Key-based Approach Towards Efficient Federated Unlearning

Rui Zhen Xu, Sheng Yi Hong, Po Wen Chi, Ming Hung Wang

研究成果: 書貢獻/報告類型會議論文篇章

摘要

Federated learning is an approach that ensures privacy in machine learning, but it has its limitations when it comes to preserving the right to be forgotten. To address this challenge, we propose a new method called Unlearning Key Revocation List (UKRL) for implementing federated unlearning. Our approach does not require clients' data or models to be unlearned; instead, we use revocation keys to remove clients from the model. We pre-trained the model to recognize these keys, so the model will forget the revoked clients when their revocation keys are applied. We conducted four experiments using MNIST datasets to verify the effectiveness of our approach, and the results showed that our work is not only effective but also time-saving since the unlearning time is 0. In conclusion, we provide a new perspective on achieving federated unlearning.

原文英語
主出版物標題Proceedings - 2023 18th Asia Joint Conference on Information Security, AsiaJCIS 2023
發行者Institute of Electrical and Electronics Engineers Inc.
頁面17-24
頁數8
ISBN(電子)9798350341638
DOIs
出版狀態已發佈 - 2023
事件18th Asia Joint Conference on Information Security, AsiaJCIS 2023 - Hybrid, Tokyo, 日本
持續時間: 2023 8月 152023 8月 16

出版系列

名字Proceedings - 2023 18th Asia Joint Conference on Information Security, AsiaJCIS 2023

會議

會議18th Asia Joint Conference on Information Security, AsiaJCIS 2023
國家/地區日本
城市Hybrid, Tokyo
期間2023/08/152023/08/16

ASJC Scopus subject areas

  • 人工智慧
  • 電腦網路與通信
  • 資訊系統
  • 安全、風險、可靠性和品質

指紋

深入研究「A Revocation Key-based Approach Towards Efficient Federated Unlearning」主題。共同形成了獨特的指紋。

引用此