2/27/2023 Aaron Seidlitz, Illinois CS
The safety and reliability of software has never been more important, and Illinois CS professor Reyhaneh Jabbarvand will utilize NSF funding to improve outcomes.
Written by Aaron Seidlitz, Illinois CS
Incredible technological developments through software impact everything from the continuously increasing capabilities of cell phones to the autonomous cars that now drive millions of miles. But there is a great responsibility associated with ensuring the safety and reliability of these enhanced technologies. Illinois Computer Science professor Reyhaneh Jabbarvand’s research in programming languages and software engineering intersects with deep learning to provide a new solution to this challenge.
She believes that recent advancement of artificial intelligence (AI) and machine learning (ML) can greatly enhance detection and root causing of bugs – specifically stated as such in her recently awarded NSF CAREER Award proposal entitled, “Advancing Neural Testing and Debugging of Software.”
“There are some challenges inherent to the way we are currently doing testing and debugging. Through this project, research will take place to rethink this process by utilizing ML to reformulate and rethink these problems,” Jabbarvand said.
The professor will work on three research goals by combining theory building, empirical data-driven research, and tool building.
According to Jabbarvand’s project proposal, the first goal is to design semantically robust neural models of code and develop systematic approaches for high-quality dataset generation. Second, she will advance several techniques to construct deep test oracles for functional and non-functional testing. And, finally, she will design interpretation techniques for extending and reusing the knowledge of neural models to unify testing and debugging.
The resulting impact, she said, can help both practitioners and researchers in the community to test and debug software smarter and faster, significantly impacting software quality.
“For example, generating test oracles – which is the ability to determine if the execution of a code produces the expected outcome or not – is one of the most challenging facets of software testing,” Jabbarvand said. “A key idea in the CAREER proposal takes apart that process. Instead of predicting the expected output, which is the main reason the oracle problem is hard and undecidable, we learn how the test and code are ‘differently correlated’ when the test result is passing compared to failing.
“Our preliminary results are promising, showing deep learning techniques are helpful to capture such correlations and associations in data, and our preliminary results show that we can achieve 93% accuracy in solving this very hard problem.”
In addition, this project will provide Jabbarvand an opportunity to work alongside students in her Intelligent CAT Lab to design tools for meaningful interpretations of the ML4Code models that influence prompt bug localization. ML4Code interpretability also enables model-in-the-loop analysis, i.e., understanding the model's limitations and improving it further with such knowledge.
Since joining the faculty at Illinois CS in 2020, Jabbarvand has drawn inspiration from her students. She also credits mentorship from colleagues as an important part of her own development.
The NSF CAREER Award represents a personal and professional achievement that provides plenty of significance.
“To me, this means that my community recognizes the validity and significance of my ideas. Personally, it indicates that I achieved one of the most important milestones you can earn as a young faculty member,” Jabbarvand said. “I would like to thank fellow CS professors Darko Marinov and Tandy Warnow; their help was instrumental in submitting this CAREER Award proposal. I'm especially thankful to have Darko as my mentor. He always challenges my ideas by asking hard questions and making me think deeper about the problems and better solutions.
“Tandy organizes helpful internal workshops and patiently goes through the CAREER proposal solicitation, explaining how a strong proposal addresses the requests in it.”