Home // Research // FSTM // DCS // News and Events // Best Paper Award at ACIIDS 2020

Best Paper Award at ACIIDS 2020

"A proof of concept to deceive humans and machines at image classification with evolutionary algorithms"
By Raluca Chitic, Nicolas Bernard and Franck Leprévost
Published in "Intelligent Information and Database Systems - 12th Asian Conference, ACIIDS 2020, Phuket, Thailand (March 23-26, 2020),
Ngoc Thanh Nguyen, Kietikul Jearanaitanakij, Ali Selamat, Bogdan Trawinski and Suphamit Chittayasothorn (Eds.). Vol. 11., p. 467-480 (2020)


The range of applications of Neural Networks encompasses image classification. However, Neural Networks are exposed to vulnerabilities, and may misclassify adversarial images, leading to potentially disastrous consequences. Our contribution is a proof of concept of a black-box, targeted, non-parametric attack using evolutionary algorithms to fool both neural networks and humans at the task of image classification. Our feasibility study is performed on VGG-16 trained on CIFAR-10. Given two categories of c1 and c2 of CIFAR-10, and an original image classified by VGG-16 as belonging to c1 we evolve this original image to a modified image that will be classified by VGG-16 as belonging to c2, although a human would still likely classify the modified image as belonging to c1.