Thomas Vogel
Search-based techniques (e.g., using genetic algorithms) are a state-of-the-art means to facilitate test case generation and program repair approaches. For a given software, such approaches typically use one static configuration of a search technique (e.g., parameters, operators, and fitness functions) to automatically produce test cases or patches. Since the performance of a search technique highly depends on its configuration, an appropriate configuration is selected once offline. We go one step further: Our hypothesis is that a static configuration of a search technique throughout the search process is not an optimal choice, as this does not take into account the current state and progress of the search. Therefore, we envision a dynamic configuration that is controlled online during the search and that outperforms uncontrolled test case generation and program repair approaches. To this end, we (1) perform an evolution analysis to learn the relevant dynamic features that impact the performance of a search technique for a given problem, (2) design and develop controllers for the evolution based on these features, and (3) incorporate humans in the loop to support the automated control.
Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - 392561203