Ethan Fast (University of Virginia)
Mutational robustness describes how likely a variant's phenotype is to remain constant in response to mutations applied within its genotype. This measure has not been evaluated generally across software, nor more specifically in the context of a genetic programming (GP) approach to automated program repair. We provide an analysis of this metric across eight benchmark programs, representing each program's genotype as an AST, and each phenotype with a regression test suite. First, we quantify the robustness of these programs, with respect to the mutational operators used in existing GP program repair methods, and analyze the extent that these results can be applied to software more broadly. Next, we evaluate the relationship between program robustness and repair success in existing GP techniques. Finally, we identify a class of program faults which inhibit robustness under current repair methods, and propose changes in variant representation under GP that enable the repair of these more difficult bugs.
Mentor: Stephanie Forrest
SFI Host: Ginger Richardson