My research focuses on understanding and assuring deep learning to ensure models can be reliably and securely deployed in the real world. In this talk, I will first discuss what deep learning assurance is and our plans for analyzing models, code and training processes to achieve the goal of assurance. I will highlight three results of our recent work: (1) empirical studies and analysis of state-of-the-art deep learning models for vulnerability detection and security; (2) dynamic analysis of training processes for early detection of malfunctioned models; and (3) numerical stability challenges in deep learning and how they may be addressed.
Wei Le is an associate professor at Iowa State University. Her research lies in the intersection of program analysis, software engineering and machine learning. Her work has appeared at top-tier venues including ICSE, FSE, ICML, ASE, ISSTA, TOSEM and TSE. She is a recipient of an NSF career award, a Google Research Award and the distinguished paper award at the top software engineering conference FSE.