DeepTheft: Stealing DNN Model Architectures through Power Side Channel - 2024

Explore a cutting-edge IEEE conference talk on DeepTheft, a novel technique for stealing Deep Neural Network (DNN) model architectures through power side channel attacks. Delve into the research presented by Garrison Gao, which uncovers potential vulnerabilities in DNN implementations. Learn about the methodology, implications, and potential countermeasures for this innovative approach to model extraction. Gain insights into the intersection of cybersecurity and machine learning, and understand the importance of protecting intellectual property in AI systems.