Optimizer Tournament for Search-Based SE: The Widely-Recommended NSGA-II Is Often Beaten Using 1/20th the Evaluations

arXiv·medium signal

Tuning modern software means searching hundreds of interacting options where scoring one setting can require a full build or test run, so optimizer choice matters — yet common guidance is misleading. Running a tournament of optimizers for search-based software engineering, the authors show NSGA-II, though widely recommended, is frequently matched by other algorithms using only 1/20th as many evaluations. For practitioners doing config tuning or search-based SE, the message is to benchmark cheaper optimizers before defaulting to the popular one.