Fig. 3From: Universal adversarial attacks on deep neural networks for medical image classificationNormalized confusion matrices for Inception V3 models attacked with targeted UPAs with \(p = 2\) on test images in skin lesions (left panels), OCT (middle panels), and chest X-ray image datasets (right panels). Further, \(\zeta = 2\%\) for skin lesions and chest X-ray image datasets, and \(\zeta = 6\%\) for OCT image datasetBack to article page