Research on Detection and Defense Methods of Adversarial Samples

Abstract:
As an important part of artificial intelligence technology, deep learning is widely used in computer vision, natural language processing and other fields. Studies have shown that the existence of adversarial attacks poses a potential threat to the secure application of deep learning models, which in turn affects the security of the model.On the basis of briefly describing the concept of adversarial samples, this paper analyzes the main ideas of adversarial attack defense, studies the representative classical adversarial sample detection methods and defense methods, and analyzes the advantages and disadvantages of various methods from the perspective of defense types and algorithm characteristics.