Dynamic adaptive Search Based Software Engineering needs fast approximate metrics (keynote)

Abstract

Abstract-Search Based Software Engineering (SBSE) uses fitness functions to guide an automated search for solutions to challenging software engineering problems. The fitness function is a form of software metric, so there is a natural and close interrelationship between software metics and SBSE. SBSE can be used as a way to experimentally validate metrics, revealing startling conflicts between metrics that purport to measure the same software attributes. SBSE also requires new forms of surrogate metrics. This topic is less well studied and, therefore, remains an interesting open problem for future work. This paper 1 overviews recent results on SBSE for experimental metric validation and discusses the open problem of fast approximate surrogate metrics for dynamic adaptive SBSE.

Search based software engineering for software product line engineering

“…Therefore, there is a requirement for thorough evaluation of the proposed tools and techniques using suitable feature models. A thorough evaluation of any search based software engineering approach requires both an experimental and an empirical dimension [57,114].…”

Deploying Search Based Software Engineering with Sapienz at Facebook

“…Such industrial applications allow us to evaluate in both laboratory and in industrial/practitioner settings [52]. As has been argued previously [32], both forms of evaluation are important in their own right and each tends to bring to light complementary evaluation findings. That is, laboratory evaluations tend to be more controlled, but less realistic, while industrial/practice evaluations tend to be more realistic, but less controlled.…”

Achievements, Open Problems and Challenges for Search Based Software Testing

“…In order to make progress on search based software testing of non-functional properties, we need to measure the nonfunctional properties of concern with sufficiently computationally efficient fitness functions. This need for efficient fitness computation may mandate the use of surrogates or approximations to the true measurement [47]. In this section, we focus on Search Based Energy Testing (SBET), for which we believe immediate progress can be made and for which there are already potential measurement approaches [42], [94], and possible surrogates [88].…”