- author = "James Callan",
- title = "Improving the Non-Functional Properties of Android Applications with Genetic Improvement",
- school = "Computer Science, University College, London",
- year = "2024",
- address = "London, UK",
- month = "19 " # mar,
- keywords = "genetic algorithms, genetic programming, genetic improvement",
-
URL = "
https://discovery.ucl.ac.uk/id/eprint/10189386/",
-
URL = "
https://discovery.ucl.ac.uk/id/eprint/10189386/2/Improving_the_Non_Functional_Properties_of_Android_Applications_with_Genetic_Improvement%20%282%29.pdf",
- size = "192 pages",
-
abstract = "There are 3.5 billion Android applications on the
Google Play Store. However, surprisingly little work
exists on automatically improving the source code of
Android apps, especially when compared to traditional
software. Genetic Improvement is a technique for
automatically improving software which has proven
successful in the past. However, its applicability in
the Android domain is yet to be explored. In this
thesis, we explore how GI can be used to improve
Android apps.
The first contribution of this thesis is an investigation into applying GI to Android with minimal changes from the standard technique, however, we achieved limited success. Next, we mined git repositories to try to find the changes that real developers make to improve the non-functional properties of applications. With what we learned in this study, we modified and successfully applied GI to improve the responsiveness of Android applications. We then moved on to multi-objective improvement, improving execution time and memory usage, but failing to improve network usage. We also provide a benchmark of tested applications that can be used to evaluate future automated improvement tools. We developed a profiler to find the most network-intensive methods to target and a novel mutation operator. Our final contribution is an adapted version of GI framework for network usage optimisation.
We found that Genetic Improvement is an effective tool for improving multiple non-functional properties of Android apps. We found that by using simulation-based testing, rather than testing variants on devices, we could make GI faster and more practical. We found that there were many opportunities for GI to more closely mimic the types of changes made by developers and that caching in particular is an effective change type. We recommend future work further explores this direction.",
- notes = "Supervisor: Justyna Petke",
