MUSE-Books: A Machine Unlearning Benchmark for Textual Data

MUSE-Books is a benchmark designed to evaluate six key properties of machine unlearning models. The benchmark focuses on textual data, specifically books, which are a common data type requiring unlearning. It was created by muse-bench and last updated on July 10, 2024.

Use Cases

• Benchmarking model performance on the six unlearning properties defined in the description
• Evaluating text models for verbatim and knowledge memorization removal
• Testing model utility preservation on data not intended for removal
• Assessing scalability of unlearning algorithms with respect to removal request size
• Measuring sustainability of models over sequential unlearning requests

Strengths

• Benchmark assesses six specific, defined properties for unlearning models
• Focuses on textual data, a common and challenging domain for unlearning
• Last updated on July 10, 2024, indicating recent maintenance

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

muse-bench

Freshness

Last updated 2024-07-10 02:02:44