Kolloquium

Am Montag, den 26. Juni 2023, um 16:00 Uhr hält

Cedric Richter
Universität Oldenburg

im Rahmen seiner beabsichtigten Dissertation einen Vortrag mit dem Titel

Automating Code Reviews with Machine Learning and Static Analyses

Der Vortrag findet hybrid statt:

A03 2-209 (Achtung! Raumänderung!)

BBB: https://meeting.uol.de/b/ced-em4-yta-4sg

Software maintenance including reviewing code, finding bugs and fixing them is often a central and
time-consuming part of the software development process.
Therefore, many automated tools have been developed to support the developer in the maintenance
process. However, the capabilities of these tools are often limited and human oversight is still required
in the form of code reviews. This PhD project contributes to the progress of automating the manual code
review process through machine learning and static analysis.
Neural bug detection is a common approach for automating code reviews with machine learning. Here,
a specialized neural network is trained to automatically detect and repair bugs. The training process
itself is human like in that the neural bug detector learns similar to a human code reviewer to distinguish
buggy and correct code by reviewing thousands of examples. Motivated by this similarity, a study was
conducted to investigate the commonalities and differences between code reviewers and neural bug detectors.
The findings of this study have led to substantial improvements in the training process and in the deployment
of neural network based bug detectors. More precisely, an approach was proposed for mining
real bug fixes at a large scale. This enabled the bug detectors to better imitate code reviewers by learning
from real code reviews. In addition, an approach for automatically validating the output of neural bug
detectors was developed that significantly reduced the number of false alarms. Together, the resulting
neural bug detector is not only more effective in reviewing code by finding a higher number of bugs but
is also more reliable by avoiding a significant number of false positives.
Betreuer: Prof. Dr. Heike Wehrheim