List your company
ResearchFormerly WizWand

Image Classification Defense on Tiny-ImageNet (CA, ASR)

48.52Clean Accuracy (CA)LAC
1.293613.554325.81538.0757Jun 18, 2026
Updated 4mo ago

Evaluation results

MethodLinks
LACQuantization=8-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0648.521.49
QVec (0.98)Quantization=8-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0648.321.53
Gaussian NoiseQuantization=8-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0648.126.53
QVec (0.98)Quantization=8-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0647.550.23
QVec (0.95)Quantization=8-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0647.480.27
EFRAPQuantization=8-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0647.161.83
LACQuantization=8-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0647.113.59
EFRAPQuantization=8-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0647.064.21
No defenseQuantization=8-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0646.8497.48
Gaussian NoiseQuantization=8-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0646.8154.73
No defenseQuantization=8-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0646.6997.95
QVec (0.95)Quantization=8-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0645.960.74
QVec (0.98)Quantization=4-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0636.631.38
QVec (0.95)Quantization=4-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0635.981.49
No defenseQuantization=4-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0632.5898.15
QVec (0.98)Quantization=4-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0632.431.69
LACQuantization=4-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0631.51.32
QVec (0.95)Quantization=4-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0630.9812.86
EFRAPQuantization=4-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0629.622.39
Gaussian NoiseQuantization=4-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0627.4925.06
LACQuantization=4-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0620.436.74
EFRAPQuantization=4-bit, Attack Scenario=Exploiting LLM, Model=MobileNetV22026.0619.5238.62
No defenseQuantization=4-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.0618.690.24
Gaussian NoiseQuantization=4-bit, Attack Scenario=Qu-Anti-zation, Model=MobileNetV22026.063.1131.31