OCAGE: An Input-Level One-Class Backdoor Detection Method Using Feature Map Extraction for DNN
Han Lin Tun, Haiyang Yu, Tian Xie, Ping Yi
Deep Neural Networks (DNNs) are vulnerable to backdoor attacks. In these attacks, an adversary inserts backdoor triggers during the training phase, causing misclassification of the model during inference stage when the triggers present. These hidden vulnerabilities pose significant risks, particularly in critical …
cn (code pays fourni par la source)
0 citations