Sunday 23 February 2025
The article discusses a new taxonomy of system-level attacks on deep learning models in autonomous vehicles. The researchers identified 19 highly relevant papers that satisfy all inclusion criteria and then tagged them with taxonomy categories, involving three assessors per paper.
The taxonomy includes 12 top-level categories and several sub-categories. It covers various types of attacks, such as evasion attacks where attackers try to deceive the system by creating fake or perturbed inputs, poisoning attacks where attackers inject malicious data into the training process, and manipulation attacks where attackers alter the environment in which the autonomous vehicle operates.
The researchers found that many of these attacks can have serious consequences for the safety of both humans and vehicles. For example, evasion attacks can cause a self-driving car to misinterpret its surroundings, leading it to crash or lose control. Poisoning attacks can compromise the training process, causing the model to learn incorrect patterns and make mistakes.
The taxonomy also highlights the importance of considering the entire system, including not just the deep learning models but also the environment in which they operate and the sensors and software that interact with them. The researchers emphasize that a comprehensive approach is necessary to ensure the safety and security of autonomous vehicles.
One of the key findings of the study is that many existing attacks on autonomous vehicles are based on manipulating the inputs or outputs of the system, rather than exploiting vulnerabilities in the model itself. This suggests that attackers may be focusing on easier targets, but it also means that the risks posed by these attacks may be higher than previously thought.
The researchers hope that their taxonomy will help to raise awareness about the threats facing autonomous vehicles and encourage further research into developing more secure and robust systems. They believe that a better understanding of these attacks can lead to the development of more effective countermeasures, which in turn can improve the safety and security of self-driving cars and other autonomous systems.
The study’s findings have significant implications for the development and deployment of autonomous vehicles, highlighting the need for more rigorous testing and evaluation of these systems. It also underscores the importance of considering the broader environment in which autonomous vehicles operate, rather than just focusing on the technology itself.
Overall, the article provides a comprehensive overview of the taxonomy of system-level attacks on deep learning models in autonomous vehicles, highlighting the risks posed by these attacks and the need for more research into developing more secure and robust systems.
Cite this article: “Taxonomy of System-Level Attacks on Deep Learning Models in Autonomous Vehicles”, The Science Archive, 2025.
Deep Learning Models, Autonomous Vehicles, System-Level Attacks, Evasion Attacks, Poisoning Attacks, Manipulation Attacks, Safety And Security, Cybersecurity, Artificial Intelligence, Machine Learning.
