EvoDuet: Bilevel Co-Evolution of Web Searching and Task Solving
What it is
EvoDuet is a bilevel optimization method that co-evolves LLM-generated solutions and web search queries to enhance scientific discovery tasks. It uses a retrieval gate to dynamically decide when to fetch new documents, reusing stored ones or proceeding without them to avoid redundant searches.
The problem it solves
LLM-based evolutionary search stalls when progress requires external knowledge, and naive web search tools often return redundant pages as solutions evolve.
What you could use it for
- Scientific discovery tasks requiring external knowledge
- Optimization problems where LLMs lack specific domain data
- Enhancing evolutionary search scaffolds like OpenEvolve
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.