Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

UnlearnDiff

Benchmarks

Task NameDataset NameSOTA ResultTrend
Safe generation against nudity promptsUnlearnDiff
ASR16.4
9
Adversarial RobustnessUnlearnDiff
Risk Ratio59.1
8
Concept Erasure RobustnessUnlearnDiff (UD)
Attack Success Rate30.28
7
Implicit Concept ErasureUnlearnDiff
ASR80
7
Text-to-Image GenerationUnlearnDiff
ASR80.9
7
Showing 5 of 5 rows