Defense Against Adversarial Ai
Bhavani Bhavani,Soundarya. S.,Tejashwini. V.,Sumitha. S.,,,,
DOI: https://doi.org/10.54216/jchci.070105
2024-01-01
Journal of Cognitive Human-Computer Interaction
Abstract:The increasing prevalence of deep learning technology has paved the way for a new era of AI-powered capabilities, promising revolutionary advancements across various societal domains such as healthcare and autonomous vehicles. Despite offering potent solutions to complex problems, the formidable power of these AI systems is accompanied by a susceptibility that malicious actors could exploit. Adversarial attacks, particularly targeting deep learning models, involve the crafting of altered inputs, often imperceptible changes to images, to deceive or undermine the functionality of the AI system. Within the domain of autonomous driving systems, adversarial attacks pose a severe risk. Envision a situation where a precisely manipulated adversarial attack targets a red traffic light sign, causing the AI system to misclassify it as an entirely unrelated object, perhaps identifying it as a bird. The potential consequences of such misclassifications underscore the serious impact that adversarial attacks can exert on the safety and dependability of autonomous vehicles. The potential repercussions of such misclassification are severe, with the risk of causing traffic accidents and posing a notable safety threat. Ensuring the resilience and security of AI technologies against adversarial threats is of utmost importance as AI continues to play a pivotal role in critical applications such as healthcare, finance, and autonomous systems. It necessitates a holistic strategy that melds advanced research, meticulous testing, and the deployment of robust security measures. This comprehensive approach is essential for fostering trust and mitigating potential harm in an ever- growing, AI-driven world.