Reproduce “A Survey of Reinforcement Learning for 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
Dong Wang, Hanmo You, Lingwei Zhu, Kaiwei Lin, Zheng Chen, Chen Yang, Junji Yu, Zan Wang and 1 more
This survey provides the first systematic mapping of Reinforcement Learning applications across the entire software engineering lifecycle. It identifies an explosive growth in publications since 2022, with a strong focus on software quality assurance tasks like test generation, and highlights the prevalent use of model-free RL algorithms while also revealing challenges in reproducibility and the underutilization of advanced RL techniques and real-world datasets.
Key insights
- Software Quality Assurance, particularly test generation, dominates RL-for-SE research, indicating an overconcentration in this area while other software engineering phases remain underexplored.
- A significant portion of RL-for-SE studies lack empirical validation and are non-replicable due to missing code or datasets, hindering real-world applicability and cumulative progress.
- There is a growing trend towards applying advanced RL algorithms and combining RL with Large Language Models, suggesting future research opportunities for more intelligent and adaptive SE systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.