Revealing Backdoors, Post-Training, in DNN Classifiers via Novel Inference on Optimized Perturbations Inducing Group Misclassification.

Zhen Xiang, David J. Miller, George Kesidis

Browse the full ICASSP paper archive.