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

Selective Unlearning on Lacuna-10 (test)

1.4Test Error (mean)

Finetune

Updated 1mo ago

Evaluation Results

Method	Links
Finetune 2023.02		1.4	0.16	0	0	0	0	57.33	3.3
Finetune 2023.02		1.43	-	0	-	0	-	-	-
Retrain 2023.02		1.5	0.08	0.33	0.47	0	0	52	2.16
Original 2023.02		1.53	-	0	-	0	-	-	-
CF-k 2023.02		1.53	-	0	-	0	-	-	-
Original 2023.02		1.57	0.24	0	0	0	0	59	2.16
CF-k 2023.02		1.57	0.12	0	0	0	0	58.33	2.49
Retrain 2023.02		1.6	-	0.67	-	0	-	-	-
SCRUB 2023.02		1.67	0.19	0	0	0	0	57.67	0.94
SCRUB+R 2023.02		1.67	0.19	0	0	0	0	57.67	0.94
EU-k 2023.02		1.83	-	0	-	0	-	-	-
Fisher 2023.02		1.87	-	0	-	0.01	-	-	-
SCRUB 2023.02		2.07	-	1.67	-	0	-	-	-
CF-k 2023.02		2.47	0.25	0	0	0	0	56	0.82
CF-k 2023.02		2.47	0.25	0	0	0	0	56	0.82
Retrain 2023.02		2.5	0.24	1.67	0.94	0	0	49.67	3.09
Retrain 2023.02		2.5	0.24	1.67	0.94	0	0	49.67	3.09
Original 2023.02		2.53	0.25	0	0	0	0	56.67	1.7
Original 2023.02		2.53	0.25	0	0	0	0	56.67	1.7
EU-k 2023.02		2.6	0	0	0	0.03	0	56	2.83
EU-k 2023.02		2.6	0	0	0	0.03	0	56	2.83
Finetune 2023.02		2.67	0.05	0	0	0	0	53.67	0.94
Finetune 2023.02		2.67	0.05	0	0	0	0	53.67	0.94
NegGrad+ 2023.02		2.77	-	8.67	-	0.4	-	-	-
SCRUB 2023.02		2.97	0.25	6	3.27	0	0	50.67	4.03
SCRUB+R 2023.02		2.97	0.25	6	3.27	0	0	50.67	4.03
SCRUB 2023.02		2.97	0.25	6	3.27	0	0	50.67	4.03
SCRUB+R 2023.02		2.97	0.25	6	3.27	0	0	50.67	4.03
NegGrad+ 2023.02		3.6	0.14	14.33	1.25	0.87	0.07	51	1.63
EU-k 2023.02		3.9	1.47	0	0	0.76	0.63	52	3.56
NegGrad+ 2023.02		4.3	0.43	12.67	3.3	0.95	0.08	54	2.16
NegGrad+ 2023.02		4.3	0.43	12.67	3.3	0.95	0.08	54	2.16
Bad-T 2023.02		4.9	-	0.67	-	1.34	-	-	-
Bad-T 2023.02		4.9	2.1	1.34	1.2	0.67	0.94	67.67	6.94
Bad-T 2023.02		7.3	2.2	3.26	1.83	0.33	0.47	67.33	3.4
Bad-T 2023.02		7.3	2.2	3.26	1.83	0.33	0.47	67.33	3.4