A Survey of Adversarial Examples and Deep Learning Based Data Hiding

Nowadays, the emergence of deep learning technology has brought breakthroughs to many fields and has been widely used in many practical scenarios. At the same time, the concept of adversarial examples are gradually known. By adding tiny disturbances to the original samples, the accuracy of the original classification depth model is successfully reduced, and the purpose of confronting deep learning is achieved. In this paper, the survey of adversarial examples and deep learning based data hiding are presented, and then the idea and possibility of combining them well are then puts forward, for providing a novel concept of data hiding based adversarial examples. In addition, this paper introduces the generation of adversarial examples and defense against them. The future research is prospected by using watermark to generate adversarial examples. Making use of the imperceptibility of data hiding, we present a novel concept of adversarial examples adding meaningful watermarks to the original image and attacking the deep neural network model.