Machine Unlearning on CIFAR-100 30% class-wise data forgetting (train/test) benchmark leaderboard · SOTA2 Research

Machine Unlearning on CIFAR-100 30% class-wise data forgetting (train/test)

99.9Utility (Accuracy, Train, Retained Data)Retraining

Evaluation Results

Method	Links
Retraining2026.04		99.9	63.8	0	0	0.833	0
IEU w/GA2026.04		99.8	64	0	0	0.833	0.001
FT2026.04		99.7	63.8	0.2	0.1	0.834	0.001
SCRUB2026.04		99.3	77	0.2	0.2	0.833	0.008
FT2026.04		99.2	75.7	0	0	0.833	0.004
RL2026.04		99.2	62.5	0	0	0.833	0.004
IEU w/GA2026.04		99	74.4	0	0	0.833	0.001
RL2026.04		98.9	75	0	0	0.833	0.003
Retraining2026.04		98.6	74.4	0	0	0.836	0
IEU w/Noisy2026.04		98.2	64.2	0	0	0.835	0.005
IEU w/Noisy2026.04		97.7	74.3	0	0	0.833	0.003
IEU w/GA+Noisy2026.04		97.4	74.2	0	0	0.833	0.003
IEU w/GA+Noisy2026.04		96.8	62.6	0	0	0.837	0.009
SCRUB2026.04		96	64.5	0	0	0.838	0.01
SALUN2026.04		89.7	69.7	1.7	1.2	0.83	0.034
SALUN2026.04		74.9	58.4	0.4	0.3	0.839	0.063