18 Nov
Artificial Intelligence is an upcoming trend in software engineering, and software testing in particular.
In this article, I summarise recent developments on research in this area.
The role of unit testing in AI for software engineering
Applications of Artificial Intelligence (AI) techniques for challenges of unit testing are gaining traction in the scientific community. Big research institutes are starting to work on this area. Besides survey on potential applications of AI techniques to software testing in 2019 (“Machine Learning Applied to Software Testing: A Systematic Mapping Study” and
AI for Testing Today and Tomorrow: Industry Perspectives), a recent (2021) book on AI for software testing dedicates 4 of its chapters (chapters 9-12) to unit testing (“Artificial Intelligence Methods for Software Engineering”), and a recent (2021) paper on artificial intelligence along the software engineering process also shows many opportunities for the area of software testing (“A Survey on Supporting the Software Engineering Process Using Artificial Intelligence”).
Available data
The success of AI models is moslty dependent on the data that can be used to train it. Recently, there has been some effort to obtain such data by crawling and annotating publicly available github repositories and make the crawling results available to the common public. The biggest effort in doing so was performed by the CrossMiner project funded by the European Union crossminer.org. For more abstract representations of “Mining Software Engineering Data from GitHub” These data collections can pave the way to training AI algorithms that can be used out of the box for new software projects, minimizing the configuration effort (and thus knowledge requried to use the algorithms). With an increased availability of data, the effort to find new solutions also becomes easier, making it possible for researchers to create AI algorithms that tackle a wide range of challenges effectively. By relying on the same set of data, these algorithms also become comparable, making it easy to decide which solution performs better for a particular problem, even if individual algorithms are developed by
different researchers.
Prioritizing Test Cases
One promising area of software testing in which AI-based techniques (more specifically,
meta-heuristic algorithms such as genetic search) have shown promising results
is the priorization of test cases “On the use of evolutionary algorithms for test case prioritization in regression testing considering requirements dependencies”.
Using so-called meta-heuristic search algorithms, the best combination of test cases can be found for a particular software under test, by trading off different conflicting goals such as minimizing execution time of tests, or maximizing code coverage and mutation scores.
Using so-called meta-heuristic search algorithms, the best combination of test cases can be found for a particular software under test, by trading off different conflicting goals such as minimizing execution time of tests, or maximizing code coverage and mutation scores.
What do you thinkg of these trends? What do you think of AI for unit testing in general? Let me know in the comments section below!
Daniel Lehner is a PhD student at the Johannes Keppler University in Linz, Austria. His research focus is on traceability of software information between different stages in the development cycle. During his master studies in Business Informatics at the Vienna University of Technology, Daniel worked as a consultant in the area of software quality assurance.
Information
Daniel Lehner