List your company
ResearchFormerly WizWand

Adversarial Attack Success Rate (ASR) on CIFAR10 (test)

99ASRMDSE
-3.9622.7749.576.23Sep 7, 2022Apr 19, 2023Nov 30, 2023Jul 12, 2024Feb 22, 2025Oct 5, 2025May 18, 2026
Updated 5mo ago

Evaluation results

MethodLinks
MDSEAttack Steps=19, Max Epsilon=0.0592022.0999——
M-DI-FGSM-EnsembleAttack Steps=19, Max Epsilon=0.0592022.0992.1——
MDSEAttack Steps=40, Max Epsilon=0.0312022.0990.4——
MDSEAttack Steps=19, Max Epsilon=0.0312022.0988.4——
M-DI-FGSM-EnsembleAttack Steps=40, Max Epsilon=0.0312022.0986.6——
M-DI-FGSM-EnsembleAttack Steps=19, Max Epsilon=0.0312022.0982.8——
SAGAAttack Steps=19, Max Epsilon=0.0592022.0960.4——
M-DI-FGSMAttack Steps=19, Max Epsilon=0.0592022.0958.9——
MoCo-EASetting=Obfuscated gradients, Norm=l_inf2026.0553.8——
Max AutoAttack Steps=19, Max Epsilon=0.0592022.0951.5——
Max PGDAttack Steps=19, Max Epsilon=0.0592022.0944.2——
MoCo-EASetting=Robust model, Norm=l_inf2026.0540.5——
AASetting=Robust model, Norm=l_inf2026.0540.3——
PGDSetting=Robust model, Norm=l_inf2026.0539.6——
AAASetting=Robust model, Norm=l_inf2026.0539.1——
MI-FGSMSetting=Robust model, Norm=l_inf2026.0538.7——
SAGAAttack Steps=19, Max Epsilon=0.0312022.0931.9——
SAGAAttack Steps=40, Max Epsilon=0.0312022.0926.6——
M-DI-FGSMAttack Steps=19, Max Epsilon=0.0312022.0924.9——
M-DI-FGSMAttack Steps=40, Max Epsilon=0.0312022.0923.3——
Max PGDAttack Steps=19, Max Epsilon=0.0312022.0916.4——
Max PGDAttack Steps=40, Max Epsilon=0.0312022.0915.4——
Max AutoAttack Steps=19, Max Epsilon=0.0312022.0915.3——
Max AutoAttack Steps=40, Max Epsilon=0.0312022.0915——
PGDSetting=Obfuscated gradients, Norm=l_inf2026.0513.5——
MI-FGSMSetting=Obfuscated gradients, Norm=l_inf2026.0513.3——
AAASetting=Obfuscated gradients, Norm=l_inf2026.0512.8——
AASetting=Obfuscated gradients, Norm=l_inf2026.050——
HARTArchitecture=ResNet18, Attack=APGD, Input Coding=Direct coding2025.03—71.4998.54
HARTArchitecture=ResNet18 (Adv. trained), Attack=APGD, Input Coding=Direct coding2025.03—21.270.74
HARTArchitecture=VGG11, Attack=APGD, Input Coding=Direct coding2025.03—54.8298.13
PDSGArchitecture=ResNet18, Attack=APGD, Input Coding=Direct coding2025.03—75.1899.97
PDSGArchitecture=ResNet18 (Adv. trained), Attack=APGD, Input Coding=Direct coding2025.03—21.5671.92
PDSGArchitecture=VGG11, Attack=APGD, Input Coding=Direct coding2025.03—45.5899.84
RGAArchitecture=ResNet18, Attack=APGD, Input Coding=Direct coding2025.03—67.0494.77
RGAArchitecture=ResNet18 (Adv. trained), Attack=APGD, Input Coding=Direct coding2025.03—17.9357.37
RGAArchitecture=VGG11, Attack=APGD, Input Coding=Direct coding2025.03—46.488.51
STBPArchitecture=ResNet18, Attack=APGD, Input Coding=Direct coding2025.03—71.3699.67
STBPArchitecture=ResNet18 (Adv. trained), Attack=APGD, Input Coding=Direct coding2025.03—14.3441.39
STBPArchitecture=VGG11, Attack=APGD, Input Coding=Direct coding2025.03—46.9699.25