Presentation at International Joint Conference on Neural Networks (IJCNN)

Topic: Not All Adversarial Examples Require a Complex Defense: Identifying Over-optimized Adversarial Examples with IQR-based Logit Thresholding