预遗忘阶段(Pre-unlearning Stage):在遗忘过程开始之前检测恶意请求或制定遗忘规则。 遗忘阶段(In-unlearning Stage):监控模型变化,如果检测到异常则停止遗忘过程。 后遗忘阶段(Post-unlearning Stage):保护遗忘后模型的信息泄露,或恢复模型到攻击前的状态。 3.3 威胁模型 (1)攻击角色(Attack Roles) 数据贡献者 (R1)...
另一方面,机器遗忘(machine unlearning, MU)是从训练好的ML模型中去除某些数据点或特征的过程,而不影响其性能[4]。MU是一个相对较新的、具有挑战性的研究领域,它关注于开发从训练有素的模型中删除敏感或不相关数据的技术。MU的目标是确保经过训练的模型不受偏见和敏感信息的影响,这些信息可能会导致负面结果[5]。
3. 对于整体 Machine Unlearning 的工作流大致分为三个部分: 《A survey of machine unlearning》 从 学习 -> 遗忘 ->验证 三个步骤实现整体的 unlearning全流程,其中: a)学习过程依旧是基于目前已知的学习和训练算法; b)遗忘过程分为 Exact unlearning 和 Approximate unlearning,遗忘的任务包括数据点、用户、特征...
Liu H, Xiong P, Zhu T, et al. A Survey on Machine Unlearning: Techniques and New Emerged Privacy Risks[J]. arXiv preprint arXiv:2406.06186, 2024. 论文提出了一个分类法,将现有的机器遗忘方法分为两个分支:面向数据的技术(data-oriented techniques)和面向模型的技术(model-oriented techniques)。总结...
Survey Targets Desiderata Unlearning Request Verification Methods Open Questions Exact Approximate Strong Weak Consistency Accuracy Verifiability Sample Class Feature Sequence Graph Client Retraining-based Attack-based Accuracy-based Relearn Time-based Theory-based Information bound-based Universality Security ...
not yet featured the use of machine unlearning but could benefit greatly from it. We hope this survey serves as a valuable resource for ML researchers and those seeking to innovate privacy technologies. Our resources are publicly available at https://github.com/tamlhp/awesome-machine-unlearning. ...
Machine Unlearning: Solutions and Challenges TETCI 2024 A Survey on Federated Unlearning: Challenges, Methods, and Future Directions arXiv 2023 A Survey of Federated Unlearning: A Taxonomy, Challenges and Future Directions arXiv 2023 Exploring the Landscape of Machine Unlearning: A Comprehensive Survey ...
然而,大多数现有方法仍存在不同层面的局限,而仍在发展中的 LLM 的部分特性则为 Machine Unlearning 带来了更多挑战。在。2023 年 10 月发布的综述《Large Language Model Unlearning》、2024 年的综述《《Digital Forgetting in Large Language Models: A Survey of Unlearning Methods》以及近期的工作中均指出了 ...
machine-unlearningdata-deletiondata-removal UpdatedJul 27, 2024 tamlhp/awesome-machine-unlearning Star675 Code Issues Pull requests Awesome Machine Unlearning (A Survey of Machine Unlearning) data-privacygdprright-to-be-forgottenmachine-unlearningcontinual-learningforgettingdata-deletionunlearningmembership-infere...
Machine Unlearning: A Comprehensive Survey As the right to be forgotten has been legislated worldwide, many studies attempt to design unlearning mechanisms to protect users' privacy when they want t... W Wang,Z Tian,C Zhang,... 被引量: 0发表: 2024年 Towards Unbounded Machine Unlearning Deep...