CloudLeak - DNN Model Extractions from Commercial MLaaS Platforms
Explore the vulnerabilities of Deep Neural Networks (DNN) deployed on commercial Machine Learning as a Service (MLaaS) platforms in this 40-minute Black Hat conference talk. Delve into the challenges of creating successful DNN models, including the need for vast amounts of data and computing power. Examine how cloud-based deep learning services have emerged as a cost-effective solution for users with limited resources. Investigate potential security risks associated with these platforms, focusing on model extraction techniques. Learn about the research conducted by Yier Jin, Honggang Yu, and Tsung-Yi Ho on DNN model extractions from MLaaS platforms. Gain insights into the implications of these vulnerabilities for both service providers and users in the rapidly evolving field of machine learning and artificial intelligence.