The University of Illinois at Urbana-Champaign
Year in School
REU Faculty Mentor
Rehyan Jabbaravand Behrouz
Research Area Interest
Programming Languages, Formal Methods, and Software Engineering
Bug Dataset Detection
Biography & Research Abstract
To use machine learning and specifically deep learning for software analysis tasks that require detecting, localizing, and repairing software bugs, the first step is to have a (1) large and (2) high-quality training dataset of the buggy and non-buggy versions of the code. There are some bug datasets such as Defects4J and BugSwarm that involve real-world bugs collected from open-source projects. However, these datasets are relatively small, i.e., each contains around 800 unique bugs. State-of-the-art relies on mutation testing and specifically, higher-order mutants to inject artificial bugs into the code. However, there is always a debate about whether artificial bugs are representative of real bugs or not. In this project, we aim to use generative models to learn how to generate bugs that mimic real-world bugs. Such techniques can help with the generation of many bugs to use for training machine learning for code analysis tasks.
I'm a rising sophomore majoring in computer science and minoring in economics. I'm interested in artificial intelligence, data science, machine learning, and statistics. I've always been interested in the mathematics portion of computer science and how we can design algorithms to influence the stock market, and the importance of data analysis in this process. I have also worked as a research assistant previously and worked on analyzing the pollster rankings in college basketball. In the fall, I'll be conducting research on optimization methods in QIS at IBM - Illinois Discovery Accelerate Institute and be a course assistant for CS 361.