Reproduce “The Weights can be Harmful: Pareto Search versus Weighted Search in Multi-Objective Search-Based Software Engineering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tao Chen, Miqing Li
An empirical study found that Pareto search, particularly NSGA-II, yields higher quality solutions than weighted search in most multi-objective Search-Based Software Engineering problems, even when explicit stakeholder preferences are known. This contradicts the long-held belief that weighted search is superior in such scenarios, demonstrating NSGA-II won in 77% of two-objective cases with sufficient budget.
Key insights
- There is no single best normalization method for weighted search across all multi-objective SBSE problems; Dynamic normalization serves as a generally robust default when experimental comparison is not feasible.
- Pareto search algorithms, particularly NSGA-II, consistently produce higher quality solutions than their best weighted search counterparts when given sufficient computational resources, challenging the 'weighted search first' principle.
- While Pareto search generally offers better solution quality, weighted search is more resource-efficient for achieving its own best possible outcome, providing a trade-off between solution quality and computational cost.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.