1 paper
Binyan Xu, Fan Yang, Xilin Dai +2
Deep Neural Networks remain inherently vulnerable to backdoor attacks. Traditional test-time defenses largely operate under the paradigm of internal diagnosis methods like model re…