Adversarial Robustness on MNIST (Image Classification)

99.64Clean AccuracyUndefended

Evaluation Results

Method
Undefended2022.02	99.64	—	98.86
AC2022.02	99.6	—	0.03
Undefended2022.02	99.57	—	99.49
NONE2022.02	99.42	—	0.03
AC2022.02	99.35	—	0.03
NAD2022.02	99.24	—	0.09
ABL2022.02	99.21	—	98.51
NAD2022.02	99.19	—	0.19
NONE2022.02	99.09	—	0.2
NAD2022.02	99.06	—	32.06
Undefended2022.02	99.05	—	98.99
ABL2022.02	99.01	—	98.81
Marksman2022.10	98.8	0.1	100
NONE2022.02	98.63	—	0.11
ABL2022.02	98.4	—	89.32
Marksman2022.10	98.3	0.6	100
SecureSplit2026.01	98	—	3
RefoolMT2022.10	97.7	1.2	34.1
AC2022.02	97.59	—	0.1
PatchMT2022.10	97.5	1.4	29.8
DP-SGD2022.02	97.17	—	0.41
WaNetMT2022.10	97	1.9	90.9
No defense2026.01	97	—	92
No defense2026.01	97	—	94
MP2026.01	97	—	62
SecureSplit2026.01	97	—	3
SecureSplit2026.01	97	—	4
SecureSplit2026.01	97	—	5
WaNetMT2022.10	96.9	2	78.4
DP-SGD2022.02	96.81	—	0.38
PatchMT2022.10	96.7	2.2	99.6
DP2026.01	96	—	72
SecureSplit2026.01	96	—	2
DP2026.01	96	—	35
ANP2026.01	95	—	62
DP2026.01	95	—	92
RefoolMT2022.10	94.2	4.7	89.3
MP2026.01	94	—	75
VFLIP2026.01	94	—	8
ANP2026.01	94	—	48
No defense2026.01	94	—	70
MP2026.01	94	—	49
ANP2026.01	94	—	17
HDBSCAN2026.01	94	—	75
DP2026.01	94	—	51
No defense2026.01	94	—	94
DP-SGD2022.02	93.02	—	1.1
HDBSCAN2026.01	93	—	59
No defense2026.01	93	—	89
HDBSCAN2026.01	92	—	74
VFLIP2026.01	92	—	7
VFLIP2026.01	92	—	11
MP2026.01	92	—	57
HDBSCAN2026.01	92	—	94
ANP2026.01	92	—	68
TrMean2026.01	91	—	72
Multi-Krum2026.01	91	—	33
DP2026.01	90	—	94
VFLIP2026.01	90	—	19
HDBSCAN2026.01	89	—	80
TrMean2026.01	89	—	64
MP2026.01	89	—	81
TrMean2026.01	88	—	63
TrMean2026.01	88	—	42
Multi-Krum2026.01	87	—	59
ANP2026.01	87	—	39
TrMean2026.01	87	—	90
VFLIP2026.01	85	—	12
Multi-Krum2026.01	84	—	65
Multi-Krum2026.01	82	—	55
Multi-Krum2026.01	82	—	52