List your company
ResearchFormerly WizWand

Untargeted white-box adversarial attack on NMNIST original (test)

1.3Baseline Error RateOriginal network
1.181.992.83.61Apr 12, 2022
Updated 8mo ago

Evaluation results

MethodLinks
Original networkNetwork structure=CONV, Attack epsilon (ε)=0.00082022.041.320
Original networkNetwork structure=CONV, Attack epsilon (ε)=0.0012022.041.333.7
Original networkNetwork structure=CONV, Attack epsilon (ε)=0.00122022.041.341
Original networkNetwork structure=CONV, Attack epsilon (ε)=0.00152022.041.349
Original networkNetwork structure=FC, Attack epsilon (ε)=0.00072022.043.721
Robust networkNetwork structure=FC, Attack epsilon (ε)=0.0007, Robust training epsilon (ε)=0.0052022.043.710
Original networkNetwork structure=FC, Attack epsilon (ε)=0.00092022.043.730.7
Robust networkNetwork structure=FC, Attack epsilon (ε)=0.0009, Robust training epsilon (ε)=0.0052022.043.712.3
Original networkNetwork structure=FC, Attack epsilon (ε)=0.00122022.043.741.3
Robust networkNetwork structure=FC, Attack epsilon (ε)=0.0012, Robust training epsilon (ε)=0.0052022.043.716.7
Original networkNetwork structure=FC, Attack epsilon (ε)=0.00142022.043.748.7
Robust networkNetwork structure=FC, Attack epsilon (ε)=0.0014, Robust training epsilon (ε)=0.0052022.043.718.3
Robust networkNetwork structure=CONV, Attack epsilon (ε)=0.0008, Robust training epsilon (ε)=0.0052022.044.37
Robust networkNetwork structure=CONV, Attack epsilon (ε)=0.001, Robust training epsilon (ε)=0.0052022.044.311
Robust networkNetwork structure=CONV, Attack epsilon (ε)=0.0012, Robust training epsilon (ε)=0.0052022.044.311
Robust networkNetwork structure=CONV, Attack epsilon (ε)=0.0015, Robust training epsilon (ε)=0.0052022.044.312