Reproduce “ResearchEVO: An End-to-End Framework for Automated Scientific Discovery and Documentation” 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
Zhe Zhao, Haibin Wen, Jiaming Ma, Jiachang Zhan, Tianyi Xu, Ye Wei, Qingfu Zhang
ResearchEVO is an integrated framework that automates scientific discovery and documentation by coupling LLM-driven algorithm evolution with literature-grounded paper generation. The system discovered human-interpretable algorithms, such as DOA-MWPM, which reduced quantum error rates by 0.4%–1.3%, and ResLRA-PINN, which improved PINN training stability, while autonomously generating complete, accurately cited scientific papers for these findings.
Key insights
- Computational instantiation of the "discover-then-explain" paradigm can generate genuinely novel, human-interpretable algorithmic mechanisms by decoupling blind, fitness-driven discovery from retrospective, literature-grounded explanation.
- The bi-dimensional co-evolution engine, optimizing both functional logic and structural architecture, allows for template-free code generation and architectural breakthroughs.
- Robust documentation is achieved through sentence-level Retrieval-Augmented Generation (RAG) and explicit anti-hallucination verification, ensuring all citations are accurately grounded in an indexed literature database.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.