Testing deep neural networks

Over the past decades, we have witnessed the emergence and rapid development of deep learning. DL has been successfully deployed in many real-life applications, including face recognition, automatic speech recognition, and autonomous driving, etc. However, due to the intrinsic vulnerability and the lack of rigorous verification, DL systems suffer from quality and security issues, such as the Alexa/Siri manipulation and the autonomous car accidents, which are introduced from both the development and deployment stages. Traditionally, testing and verification are applied to improve the quality of software systems, either to find defects or to prove they are bug-free. However, due to the fundamentally different programming paradigm and logic representation from traditional software, existing quality assurance techniques can hardly be directly applied to DL systems. In this project, you are able to learn about the state-of-the-art analysis methods of deep learning and investigate how they perform on different types of neural networks.