MUFAC, MUCAC

本研究提出了两个机器遗忘基准数据集MUFAC和MUCAC，旨在评估机器遗忘算法在面部识别系统中忘记个人身份的能力。MUFAC包含超过13,000张亚洲面部图像，用于年龄分类任务；MUCAC则包含30,000张名人面部图像，用于多标签分类任务。两个数据集均包含个人身份标签，有助于评估机器遗忘算法的鲁棒性。数据集创建过程中，确保了个人身份信息在不同数据子集间的分离，以模拟真实世界场景。这些数据集主要应用于评估机器遗忘算法在保护个人隐私方面的效果，特别是在需要删除特定个人身份信息的场景中。

This study presents two machine unlearning benchmark datasets, MUFAC and MUCAC, designed to evaluate the capacity of machine unlearning algorithms to erase individual identities in facial recognition systems. MUFAC comprises over 13,000 Asian facial images for age classification tasks, while MUCAC includes 30,000 celebrity facial images for multi-label classification tasks. Both datasets are annotated with individual identity labels, which enable the assessment of the robustness of machine unlearning algorithms. During the dataset development process, the separation of individual identity information across distinct data subsets was ensured to simulate real-world scenarios. These datasets are primarily utilized to evaluate the performance of machine unlearning algorithms in safeguarding personal privacy, particularly in scenarios where specific individual identity information requires erasure.