Repairing Code with Deep Learning

Although generative models of code, like Github Copilot, are all the rage, writing code is just a small part of software development. Developers spend much more of their time maintaining code. In this talk, Miltos Allamanis will discuss deep learning models that find and fix seemingly simple but hard-to-find bugs. Specifically, these models target bugs where there is a mismatch between the (latent) intent of the developer and the source code. This requires models that reason over highly-structured data and code’s formal semantics. Here, structured deep learning models achieve state-of-the-art performance and trained models uncover previously unknown bugs in open-source projects on Github. Miltos Allamanis will conclude by discussing the open challenges in this area.

Bio: Miltos Allamanis is a principal researcher at Microsoft Research, Cambridge UK, working at the intersection of machine learning, programming languages, and software engineering. His research combines the rich structure of programming languages with deep learning to create better tools for developers, while using problems in this area to motivate machine learning research. He obtained his PhD from the University of Edinburgh. More information about him and his publications can be found at https://miltos.allamanis.com