Reproduce “Deliberate Evolution: Agentic Reasoning for Sample-Efficient Symbolic Regression with LLMs” 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
Xinyu Pang, Zhanke Zhou, Xuan Li, Fangrui Lv, Shanshan Wei, Sen Cui, Bo Han, Changshui Zhang
Deliberate Evolution (DE) is an agentic framework for Symbolic Regression that improves sample efficiency and accuracy by decoupling symbolic generation from search control. DE integrates adaptive evolutionary operators, diagnostic tools, and reflective memory, achieving higher accuracy, better generalization, and robustness on scientific benchmarks while using 60% fewer LLM evaluations than prior methods.
Key insights
- Explicitly separating symbolic generation from search control, and providing structured directional, diagnostic, and historical guidance, significantly enhances the efficiency and effectiveness of LLM-based symbolic regression.
- Transforming scalar error into localized, actionable diagnostic reports allows LLMs to make targeted structural edits, moving beyond generic trial-and-error.
- Accumulating and reflecting on past search trajectories (successes and failures) enables the system to learn and reuse experience, avoiding redundant exploration and accelerating discovery.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.