Machine Unlearning on CIFAR10 N=500 (RANDOM) benchmark leaderboard · SOTA2 Research

Machine Unlearning on CIFAR10 N=500 (RANDOM)

0KL Divergence (Train)pθr

Evaluation Results

Method	Links
pθr2026.05		0	—	0	89.7	90	100
pθr2026.05		0	—	0	96.9	97.2	100
pθ2026.05		0.01	—	0.12	96.9	100	—
LDA2026.05		0.02	0.02	0.09	96.8	100	1.3
LDA-Mix2026.05		0.02	0.02	0.05	96.8	98.7	1.6
LDA-2C2026.05		0.02	0.02	0.09	96.8	99.2	1.3
DIR-2C2026.05		0.02	0.02	0.1	96.8	100	1.3
SCRUB2026.05		0.02	0.02	0.11	96.8	99.4	0.66
DIR2026.05		0.03	0.04	0.1	96.4	99.5	1.2
SalUn2026.05		0.03	0.9	0.09	96.5	100	0.49
RL+FT2026.05		0.03	0.24	0.11	96.5	100	0.64
GA+FT2026.05		0.03	0.03	0.05	96.7	99.1	19
FT2026.05		0.03	0.03	0.08	96.7	99.7	19
GA2026.05		0.03	27	0.03	96.7	97.9	0.21
DIR2026.05		0.41	0.4	0.76	88.9	90.6	4.7
pθ2026.05		0.42	—	0.95	89.6	100	—
LDA2026.05		0.42	0.43	0.94	89.5	100	7
LDA-Mix2026.05		0.42	0.42	0.88	89.5	100	7
LDA-2C2026.05		0.42	0.43	0.94	89.6	99.9	6.9
DIR-2C2026.05		0.42	0.42	0.87	89.5	100	4.6
SCRUB2026.05		0.42	0.46	0.92	89.5	99.6	1.7
GA2026.05		0.42	0.62	0.91	89.5	99.5	0.48
RL+FT2026.05		0.43	0.47	0.83	89.1	97.7	0.79
SalUn2026.05		0.44	0.64	0.83	88.9	96.5	1.2
GA+FT2026.05		0.52	0.66	1.1	89.4	98.9	70
FT2026.05		0.54	0.67	1.1	89.5	100	70