Reproduce “EvoDuet: Bilevel Co-Evolution of Web Searching and Task Solving for Scientific 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
Young-Jun Lee, Jinheon Baek, Soyeong Jeong, Minki Kang, Seungyeon Jwa, Jonghyun Choi, Seungho Han, Dongyeop Kang
EvoDuet couples evolutionary program search with an evolving retrieval loop that adapts queries to changing knowledge gaps and ranks documents by predicted task scores. Across 21 tasks it raises mean best Normalized Discovery Gain by 3.9 and 21.0 percentage points for two models at one candidate per iteration, but falls 14.4 points with Qwen3.5-9B, showing model-dependent results.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.