Apply

Automatic Program Repair with Machine Learning

 

Rafael-Michael Karampatsis

My research focuses on automatic fault localization and program repair with deep learning. The project tackles the vocabulary and scalability issues of neural language models for source code. It is also concerned with addressing the lack of datasets for real world SStuBs (Simple Stupid Bugs) in source code, which has resulted in the use of synthetic data for evaluating fault localization techniques. 

SStuBs consist of small semantics bugs which are syntactically correct and thus hard for a developer to manually spot. Lastly, the project's main goal is to build an end-to-end system for automatically repairing SStuBs.

 

Supervisors: Charles Sutton & Mirella Lapata