Watch a 13-minute conference presentation from USENIX Security '24 exploring the systematic analysis of physical side-channel attacks on Deep Neural Networks (DNNs). Examine how researchers from Radboud University developed a taxonomy and framework for understanding various attack methodologies that can compromise DNN implementations on hardware accelerators. Learn about the relationships between threat models, attack objectives, and analysis methods, while discovering practical limitations validated through experiments on commercial DNN accelerators. Gain insights into the security implications for DNN intellectual property protection and user data privacy, along with identified challenges and proposed future research directions in this critical area of cybersecurity.