Decompiling Deep Neural Network Compiled Binary

Explore the groundbreaking DnD (Decompiling Deep Neural Network) technique in this 32-minute Black Hat conference talk. Delve into the world of Deep Neural Networks (DNNs) and their increasing prevalence in edge devices and embedded systems. Discover how DnD, the first ISA- and compiler-agnostic DNN decompiler, addresses the challenge of recovering high-level representations of DNN models from compiled binary code. Learn about the implications for security applications such as model extraction, white-box adversarial sample generation, and model patching. Gain insights from experts Antonio Bianchi, Taegyu Kim, Dave (Jing) Tian, Ruoyu Wu, and Dongyan Xu as they present their innovative approach to decompiling DNN binaries, opening new possibilities for analyzing and understanding compiled neural network models.