Reinforcement Learning for Mutation Operator Selection in Automated Program Repair

Created by W.Langdon from gp-bibliography.bib Revision:1.7656

@Misc{hanna2023reinforcement,

• author = "Carol Hanna and Aymeric Blot and Justyna Petke",
• title = "Reinforcement Learning for Mutation Operator Selection in Automated Program Repair",
• howpublished = "arXiv",
• year = "2023",
• month = "9 " # jun,
• keywords = "genetic algorithms, genetic programming, genetic improvement, APR, JaRFly, Defects4J",
• eprint = "2306.05792",
• archiveprefix = "arXiv",
• primaryclass = "cs.SE",
• URL = "https://arxiv.org/abs/2306.05792",
• size = "12 pages",
• abstract = "Automated program repair techniques aim to aid software developers with the challenging task of fixing bugs. In heuristic-based program repair, a search space of program variants is created by applying mutation operations on the source code to find potential patches for bugs. Most commonly, every selection of a mutation operator during search is performed uniformly at random. The inefficiency of this critical step in the search creates many variants that do not compile or break intended functionality, wasting considerable resources as a result. we address this issue and propose a reinforcement learning-based approach to optimise the selection of mutation operators in heuristic-based program repair. Our solution is programming language, granularity-level, and search strategy agnostic and allows for easy augmentation into existing heuristic-based repair tools. We conduct extensive experimentation on four operator selection techniques, two reward types, two credit assignment strategies, two integration methods, and three sets of mutation operators using 22300 independent repair attempts. We evaluate our approach on 353 real-world bugs from the Defects4J benchmark. Results show that the epsilon-greedy multi-armed bandit algorithm with average credit assignment is best for mutation operator selection. Our approach exhibits a 17.3percent improvement upon the baseline, by generating patches for 9 additional bugs for a total of 61 patched bugs in the Defects4J benchmark.",

}

Genetic Programming entries for Carol Hanna Aymeric Blot Justyna Petke

Citations