[4Rin1] Interactive 2

[4Rin1-14] An Attempt of Coevolutionary Generation of Adversarial Examples for Image Classifier

〇Kosuke Nakanishi¹, Takahiro Suzuki¹, Yuki Nakashima¹, Satoshi Ono¹ (1.Kagoshima University)

Keywords:Adversarial Examples, Coevolution, Evolutionary Computation, Two-Dimensional-Discrete Cosine Transform, Deep Neural Network

In recent years, deep learning, which plays a central role in the field of machine learning, has shown excellent performance in applications such as image recognition and speech recognition. On the other hand, recent studies have shown that vulnerabilities are affected by hostile samples (Adversarial Examples: AE) generated by attackers to induce misclassification. To use machine learning technology safely in the real world, various AEs can be generated and vulnerability of the classifier can be investigated. In this study, we propose a method for generating AEs using a symbiotic evolution algorithm, one of the co-evolution models. In this method, the convergence speed of the AE generation method of the previous research to the optimal solution is improved.