Master Thesis in Software Repair

1. Building a repair benchmark for Android apps
The objective is to create a dataset of Android apps associated with their bug revealing test cases (e.g., input sequences of Monkey or GrodDroid that reveal bugs).

2. Checking patch correctness via checking patch description against test behaviour
The objective here is to devise a technique for filtering out automatically-generated patches that are incorrect by assessing whether the code change actions are semantically relevant to the failing test cases that they resolve.

3. Building a taxonomy of conversion errors when transforming CPU-intended machine learning code into GPU-intended machine learning code
The objective here is to study and characterise the variety of faults that automatic converters for machine learning code can make when converting code to adapt it for GPU architectures. Eventually, we can build tool support to automate transformations.