Effective Universal Unrestricted Adversarial Attacks Using a MOE Approach

Recent studies have shown that Deep Leaning models are susceptible to adversarial examples, which data, in general images, intentionally modified fool a machine learning classifier. In this paper, we present multi-objective nested evolutionary algorithm generate universal unrestricted examples black-box scenario. The attacks performed through the application of well-known image filters available several processing libraries, modern cameras, and mobile applications. optimization takes into account not only attack success rate but also detection rate. Experimental results showed approach is able create sequence capable generating very effective undetectable attacks.