报告题目：Deep Learning for Practical Program Analysis
报告人：Shiqi Shen, National University of Singapore (沈诗琦，新加坡国立大学)
Shiqi Shen is a Ph.D. student in the School of Computing at National University of Singapore (NUS). Her research interests include software security, program analysis and security in machine learning. During her Ph.D., she has published many high-quality peer-reviewed research papers in well-regarded security conference proceedings (e.g., CCS, Usenix and NDSS).
Program analysis is a classical problem in computer security. It analyzes the behaviors of software which are essential for multiple security applications such as hardening, bug-finding, clone detection and program repair. However, developing practical program analysis techniques that scale to real-world programs and automatically adapt to a given platform/language is challenging. To address these challenges, I investigate an alternative line of research, which utilizes the existing advance in machine learning to enhance state-of-the-art program analysis techniques. To show the effectiveness and generality of machine learning approaches, I evaluate it on two different scenarios: type recovery and symbolic execution. The evaluation results demonstrate that the machine learning approaches significantly improve the existing techniques in terms of efficiency, accuracy and adaptability.