Automated Program Repair

This is a graduate-level look at how machines can debug and fix buggy code on their own — the core question is how to localize a fault automatically and then synthesize a patch that actually holds up, not just one that happens to pass the existing tests. You'll work through the four main repair families (heuristic, template, constraint-based, and learning-based) and spend most of the semester on a project, with a midterm checkpoint and final presentations driving the timeline. It sits at the intersection of software engineering, program analysis, and increasingly ML, and matters because patch quality and validation are exactly where modern AI coding tools still struggle.