Review of Adversarial Security Threats to Electromagnetic Spectrum Artificial Intelligence Models
Electromagnetic spectrum plays a crucial role in modem society,forming the foundation of national strategic resources and providing key support in fields such as communication,navigation,scientific research,and national defense.To address myriad challenges in efficient management and utilization of electromagnetic spectrum,Artificial Intelligence(AI)technologies are widely applied at the physical layer.However,studies have found that AI models'reliance on data makes them vulnerable to malicious attacks during both training and testing phases.To advance research on attacks and defenses related to AI models of electromagnetic spectrum,and to ensure secure application of AI models,thereby enhancing electromagnetic security capabilities,this paper reviews adversarial attack methods on AI models of physical layer.This includes principles and methods of attacks during training and testing phases.A review is given from the perspectives of data,model,and specific characteristics of the electromagnetic signal domain.Finally,the paper identifies three potential research directions attacks,evaluation,and system development,and concludes with a summary.
万方数据
维普
