A Capable Crossover Technique on Automatic Program Repair
Kou Ryotaro, Yoshiki Higo, Shinji Kusumoto · 2016
In software development, debugging is indispensable to guarantee the reliability. However, debugging is becoming more difficult because software is becoming larger and more complex. Thus, techniques for supporting debugging, especially automatic program repair techniques based on genetic programming distinguish themselves due to their capability. Genetic programming produces many modified programs by three operations: selection, mutation, and crossover. In this research, we focus on crossover that can bring a large modification at one operation, and we are conducting research on capable crossover by selecting several modified programs not randomly but using the guidance indicating their properties. In this paper, we propose a new crossover technique based on comparing modified programs' properties.