A survey of genetic improvement search spaces

Abstract

Genetic Improvement (GI) uses automated search to improve existing software. Most GI work has focused on empirical studies that successfully apply GI to improve software's running time, fix bugs, add new features, etc. There has been little research into why GI has been so successful. For example, genetic programming has been the most commonly applied search algorithm in GI. Is genetic programming the best choice for GI? Initial attempts to answer this question have explored GI's mutation search space. This paper summarises the work published on this question to date. CCS CONCEPTS • Software and its engineering → Search-based software engineering;

Empirical Comparison of Search Heuristics for Genetic Improvement of Software

“…In future work, we intend to better study the fitness landscape of GI scenarios. Indeed, little is known about them, with a short survey published only last year [21]. If our results showed evidence of the possible improvements a more suitable search process can yield, there is no doubt fitness landscape analysis will lead to a deeper understanding of what evolutionary process should be preferred for GI.…”

“…In order to come closer to answering the question which search strategy is the most efficient and effective in GI, a few studies tried to answer the question of what the search space of code changes looks like in practice [21]. In each work a particular GI framework was used, and none compared the different possible search strategies.…”

“…That case study suggested that all surveyed approaches were able to produce improved software versions, although resulted in no clear winner. Finally, there have been studies directly investigating the program search space of GI problems [21], for example reporting on large number of neutral variants [28]. Rather than primarily relying on empirical observations, these studies can bring better insight the search process and indirectly hint towards which type of search should be preferred.…”

Synthetic Benchmarks for Genetic Improvement

“…In general, synthetic models should ultimately reflect known GI search spaces [6] and include features that can have a strong impact on GI approaches (e.g., mutational robustness [8], or plastic regions [3]). However, models should also compromise between high complexity and fidelity, and overall speed usage.…”

Injecting Shortcuts for Faster Running Java Code

“…Genetic Improvement of software [1] (GI) is a rapidly expanding area of research. GI uses automated search to improve existing software.…”