Machine Unlearning on WMDP Cyber benchmark leaderboard · SOTA2 Research

Machine Unlearning on WMDP Cyber

7.19RelTRU

Evaluation Results

Method	Links
TRU2026.02		7.19	8.81	9.17	6.46	2.6	2.67
PO2026.02		1.92	3.76	0.1	2.95	1.1	0.66
NPO2026.02		1.18	0.01	0	0.01	0	0
WGA2026.02		1.11	0.04	0.06	2.75	2.13	1.73
Base2026.02		0.86	0.04	0.19	5.53	3.45	3.64
RMU2026.02		0.49	0.04	0.05	4.02	3.57	3.11
GradAscent2026.02		0	0	0	0	0	0
GradDiff2026.02		0	0	0	0	0	0
KL2026.02		0	0	0	0	0	0