muse-bench/MUSE-Books|机器学习数据集|数据隐私数据集

MUSE-Books是一个全面的机器遗忘评估基准，旨在评估遗忘模型的六个关键属性：无逐字记忆、无知识记忆、无隐私泄露、对非移除数据的效用保持、移除请求大小的可扩展性以及顺序遗忘请求的可持续性。该数据集包含J.K.罗琳所写的《哈利·波特》系列书籍，分为五个子集：`raw`、`verbmem`、`knowmem`、`privleak`和`train`。每个子集都有特定的用途和分割，如`verbmem`用于评估逐字记忆，`knowmem`用于评估知识记忆和效用保持。

MUSE-Books is a comprehensive machine unlearning evaluation benchmark that assesses six key properties for unlearned models: no verbatim memorization, no knowledge memorization, no privacy leakage, utility preservation on data not intended for removal, scalability with respect to the size of removal requests, and sustainability over sequential unlearning requests. The dataset comprises the Harry Potter book series written by J. K. Rowling and is divided into five subsets: `raw`, `verbmem`, `knowmem`, `privleak`, and `train`. Each subset serves specific purposes and is partitioned accordingly, such as `verbmem` for evaluating verbatim memorization and `knowmem` for assessing knowledge memorization and utility preservation.

原始信息汇总

数据集概述

数据集名称

MUSE-Books

数据集描述

MUSE-Books 是一个综合的机器遗忘评估基准，包含五个子集：raw、verbmem、knowmem、privleak 和 train。每个子集包含不同的数据文件和特征，用于评估机器遗忘模型的不同属性。

子集详情

子集：raw

• 特征：
  • text: string
• 分割：
  • retain2: 1969626 bytes, 13 examples
  • forget: 4096855 bytes, 4 examples
  • retain1: 836924 bytes, 12 examples
  • holdout: 2328993 bytes, 3 examples
• 下载大小：5386338 bytes
• 数据集大小：9232398 bytes

子集：verbmem

• 特征：
  • prompt: string
  • gt: string
• 分割：
  • forget: 417969 bytes, 100 examples
• 下载大小：283251 bytes
• 数据集大小：417969 bytes

子集：knowmem

• 特征：
  • answer: string
  • question: string
• 分割：
  • retain_qa_icl: 1143 bytes, 10 examples
  • retain_qa: 9398 bytes, 100 examples
  • forget_qa: 9896 bytes, 100 examples
  • forget_qa_icl: 1033 bytes, 10 examples
• 下载大小：21229 bytes
• 数据集大小：21470 bytes

子集：privleak

• 特征：
  • text: string
• 分割：
  • retain: 780073 bytes, 100 examples
  • forget: 744007 bytes, 100 examples
  • holdout: 706326 bytes, 100 examples
• 下载大小：1329454 bytes
• 数据集大小：2230406 bytes

子集：train

• 特征：
  • text: string
• 分割：
  • retain2: 1969626 bytes, 13 examples
  • forget: 4096855 bytes, 4 examples
  • retain1: 836924 bytes, 12 examples
• 下载大小：3997041 bytes
• 数据集大小：6903405 bytes

数据文件路径

• knowmem：
  • retain_qa_icl: knowmem/retain_qa_icl-*
  • retain_qa: knowmem/retain_qa-*
  • forget_qa: knowmem/forget_qa-*
  • forget_qa_icl: knowmem/forget_qa_icl-*
• privleak：
  • retain: privleak/retain-*
  • forget: privleak/forget-*
  • holdout: privleak/holdout-*
• raw：
  • retain2: raw/retain2-*
  • forget: raw/forget-*
  • retain1: raw/retain1-*
  • holdout: raw/holdout-*
• train：
  • retain2: train/retain2-*
  • forget: train/forget-*
  • retain1: train/retain1-*
• verbmem：
  • forget: verbmem/forget-*