Reproduce “MLEvolve: A Self-Evolving Framework for Automated Machine Learning Algorithm Discovery” 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
Shangheng Du, Xiangchao Yan, Jinxin Shi, Zongsheng Cao, Shiyang Feng, Zichen Liang, Boyuan Sun, Xu Guo and 8 more
MLEvolve improves end-to-end machine-learning pipelines through repeated runs, measurements, and revisions under a limited budget. Its Progressive Monte Carlo Graph Search, Retrospective Memory, and adaptive code generation let candidate pipelines share past outcomes and shift effort toward promising directions. On MLE-Bench’s 75 competitions, it earns medals on 65.3% of tasks with Gemini-3.1-Pro-preview under 12 hours; the comparison uses different models and runtimes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.