Saliency Diversified Deep Ensemble for Robustness to Adversaries
https://arxiv.org/pdf/2112.03615
The idea is to promote saliency map diversity (SMD) on ensemble members to prevent the attacker from targeting all ensemble members at once by introducing an additional term in our learning objective.
通过异化显著性来增强集成模型的多样性
Thus with our approach, we try to minimize possible shared sensitivity across the ensemble members that might be exploited as vulnerability, which is in contrast to (Yang et al. 2020) who try to minimize shared ’week’ features across the ensemble members.