Generation Method of Error-Specific Adversarial Examples Using Gradient Information for the Target Class

With the advancement of AI technology, vulnerabilities of AI systems have been pointed out. Adversarial Examples (AEs), in which makes AI wrong decisions, are one of the dreaded attacks for AI. Therefore, a thorough investigation of AEs is essential for the safe use of AI. In this paper, we propose a method for generating adversarial examples using gradient information for the target class of input images. We experimentally prove that the proposed method can generate target AEs that misclassify to an arbitrary class with high probability.