Elizabeth Dinella
Neural Inference of Program Specifications
Research Abstract:
In recent years, deep learning techniques have achieved remarkable advancements across various domains, revolutionizing many fields of computer science. Inspired by these breakthroughs, my research aims to leverage the power of deep learning in the field of program analysis. Productively writing correct and secure code for software projects is a significant challenge. As such, program analysis has been an active research area for many decades. Many fruitful techniques based on rules and formal logic have emerged. Despite their successes, these approaches have some noteworthy limitations. Ultimately, my research endeavors aim to bridge the gap between deep learning and program analysis, harnessing the potential of these cutting-edge techniques to advance the state-of-the-art in software development, ensure code correctness, and bolster the security of software systems.
Bio:
Elizabeth is a PhD candidate at the University of Pennsylvania, graduating in December 2024. She specializes in machine learning for program reasoning tasks. Her research focuses on developing and utilizing models of code to improve software reliability and security. During her time as an intern at Microsoft Research, Elizabeth contributed to the DeepMerge and TOGA projects. When she's not working, Elizabeth enjoys spending time with her chow chow, Cinnabon.