A black-box adversarial attack strategy with adjustable sparsity and generalizability for deep image classifiers | Publicación