Backdoor Defense on CIFAR-10 0.1% clean data (50 images) (train) benchmark leaderboard · SOTA2 Research

Backdoor Defense on CIFAR-10 0.1% clean data (50 images) (train)

93.73ACC (Badnets)Before (No Defense)

Evaluation Results

Method
Before (No Defense)2021.10	93.73	94.82	93.89	94.1	93.78	93.64	99.97	100	98.49	92.88	99.94	94.26
No Defense2021.10	93.73	94.82	93.89	94.1	93.78	93.64	99.97	100	98.49	92.88	99.94	94.26
FP2021.10	90.69	91.29	90.74	90.26	89.27	89.96	30.71	99.95	30.82	42.38	1.35	0.13
Fine-pruning (FP)2021.10	90.69	91.29	90.74	90.26	89.27	89.96	30.71	99.95	30.82	42.38	1.35	0.13
FT2021.10	88.9	91.21	88.17	84.21	91.1	90	82.58	100	1.24	14.69	50.25	12.59
Fine-tuning (FT)2021.10	88.9	91.21	88.17	84.21	91.1	90	82.58	100	1.24	14.69	50.25	12.59
ANP2021.10	86.62	88.12	88.54	85.66	86.73	89.54	0.45	5.7	1.84	1.43	0.23	2.85
Adversarial Neuron Pruning (ANP)2021.10	86.62	88.12	88.54	85.66	86.73	89.54	0.45	5.7	1.84	1.43	0.23	2.85
FT2021.10	82.06	86.79	75.28	60.45	88.11	84.37	16.27	99.99	2.29	3.78	22.71	0.07
Fine-tuning (FT)2021.10	82.06	86.79	75.28	60.45	88.11	84.37	16.27	99.99	2.29	3.78	22.71	0.07
MCR2021.10	80.21	81.35	79.1	82.53	77.33	85.88	33.7	43.57	52.23	53.17	12.77	1.1
Model Cleanse and Repair (MCR)2021.10	80.21	81.35	79.1	82.53	77.33	85.88	33.7	43.57	52.23	53.17	12.77	1.1