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

MMA

Benchmarks

Task NameDataset NameSOTA ResultTrend
Machine UnlearningMMA (Target)
Nudity Generation Rate3.5
24
Text-to-Image GenerationMMA
ASR76
21
Defense against adversarial promptsMMA (test)
Attack Success Rate0.4
15
Implicit Concept ErasureMMA
MMA63
14
Concept ErasureMMA
ASR (%)38
14
Machine UnlearningMMA (Adversarial)
Nudity Generation Rate4.9
12
Common RobustnessMMA
ASR1.03
12
Nudity UnlearningMMA
ESD75.78
10
Adversarial RobustnessMMA
Risk Ratio28.3
8
Nudity ErasureMMA
Unsafe Rate1.7
7
Concept ErasureMMA
Nudity Rate3.2
7
Harmful prompt detectionadv-MMA
Precision100
6
Harmful prompt detectionMMA
Precision98
6
Concept UnlearningMMA
Score96.8
6
Nudity revivalMMA 1000 prompts
NSFW Image Count57
5
NSFW RemovalMMA
MMA94.12
4
Text-to-Image Jailbreak AttackMMA
Attack Success Rate (Paper)54.2
1
Showing 17 of 17 rows