Seminar Series - Automatic Software Repair Using Genetic Programming

Speaker: Dr. Claire Le Goues, University of Virginia
Date: Friday, September 21, 2012
Time: 11:15am-12:30pm
Location: 2150 Torgersen

Software quality is a pernicious problem. Mature software projects ship with both known and unknown bugs, because the number of defects typically exceeds the resources available to address them. Although there exist many tools to help with bug triage, localization, validation and confirmation, actual bug repair remains a predominantly manual, and thus incredibly expensive, process.

I will describe GenProg, a tool that uses evolutionary computation to automatically repair real bugs in legacy, off-the-shelf software. I will then present experimental evidence that GenProg is scalable and human-competitive, in that it can automatically fix a large proportion of real-world bugs quickly and cheaply. These experiments evaluate GenProg on 105 historical defects from 8 open-source programs totaling 5.1 million lines of code. Using publicly-available cloud resources, GenProg automatically repairs 60 of those 105 defects in 96 minutes for less than $8 each, on average.

Claire Le Goues is a Ph.D. candidate in Computer Science at the University of Virginia, where she will graduate in May, 2013. She received a BA in Computer Science from Harvard University in 2006, and an MS in Computer Science from the University of Virginia in 2009. Her research interests lie in the intersection of software engineering and programming languages, with a particular focus on software quality and automated error repair.