A Review on Artificial Intelligence Based Active Defense System against Online Service Attacks: A Comprehensive Analysis of Adversarial Attack and Defense in AI Systems

Abstract

Artificial intelligence technologies have been increasingly popular in recent years, with applications in computer vision, natural language processing, automated driving, and other disciplines. Artificial intelligence systems, on the other hand, are subject to adversarial assaults, which restricts the use of AI technology in critical security areas. As a result, strengthening the resilience of AI systems against adversarial assaults has become increasingly essential in AI development.The goal of this paper is to provide a thorough overview of recent research on adversarial attack and defensive methods in deep learning. This article explains adversarial attack strategies in the training and testing stages of the target model, according to the distinct stages where the adversarial assault happened. The applications of adversarial attack technologies are then sorted out. Computer vision, natural language processing, cyberspace security, and the physical environment are all areas where researchers are working. Finally, we divide the known adversarial defensive strategies into three categories data modification, model modification, and the use of auxiliary tools.