MU-Bench
收藏资源简介:
MU-Bench是由麻省大学洛厄尔分校开发的全面机器遗忘基准数据集,包含9个公开数据集,覆盖多种任务和数据模态,如图像、文本、音频和视频。数据集旨在评估机器遗忘算法在删除特定训练样本时的效果,特别关注敏感信息和过时知识的移除。创建过程中,数据集统一了删除样本和训练模型的选择,以确保公平比较。MU-Bench的应用领域广泛,包括但不限于隐私保护、数据更新和模型安全性提升。
MU-Bench is a comprehensive machine unlearning benchmark dataset developed by the University of Massachusetts Lowell. It includes nine publicly available datasets covering diverse tasks and data modalities such as images, text, audio and video. The dataset aims to evaluate the performance of machine unlearning algorithms when removing specific training samples, with particular focus on the removal of sensitive information and outdated knowledge. During its development, the dataset standardized the selection of deleted samples and trained models to ensure fair comparisons. MU-Bench has wide application fields, including but not limited to privacy protection, data updating and model security enhancement.

- 1MU-Bench: A Multitask Multimodal Benchmark for Machine Unlearning麻省大学洛厄尔分校 · 2024年



