State-of-the-art deep classifiers are vulnerable to universaladversarial perturbations: single disturbances of small magnitude that lead tomisclassification of most inputs . This phenomena may potentially result in aserious security problem . Despite extensive research in this area, there is a lack of theoretical understanding of the structure of these perturbation . We propose to useTuring patterns, generated by cellular automata, as universal perturations, to use . They significantly degrade the performance of deeplearning models . We found this method to be a fast and efficient way to create a data-agnostic quasi-imperceptible perturbantation in the black-box scenario. We find this method is a fast, efficient way of creating this method. We found

Author(s) : Nurislam Tursynbek, Ilya Vilkoviskiy, Maria Sindeeva, Ivan Oseledets

Links : PDF - Abstract

Code :


Keywords : method - patterns - fast - cellular - automata -

Leave a Reply

Your email address will not be published. Required fields are marked *