Reproduce “DEEPMED: Building a Medical DeepResearch Agent via Multi-hop Med-Search Data and Turn-Controlled Agentic Training & Inference” 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

Zihan Wang, Hao Wang, Shi Feng, Xiaocui Yang, Daling Wang, Yiqun Zhang, Jinghao Lin, Haihua Yang and 1 more

DEEPMED introduces a medical deep research agent that enhances Large Language Models' diagnostic reasoning by integrating controlled web search capabilities. It achieves state-of-the-art performance on six out of seven medical benchmarks, mitigating hallucinations and inefficient tool use.

Key insights

  • Specialized multi-hop medical search data is crucial for tailoring general DR models to medical contexts, significantly improving diagnostic capabilities beyond generic datasets.
  • Effective integration of search capabilities with medical reasoning necessitates mechanisms to control tool use, preventing context rot and redundant information retrieval in sensitive medical tasks.
  • The DeepResearch paradigm can enhance medical LLMs by grounding reasoning in verifiable external evidence, thereby reducing hallucinations and enabling evidence-based self-correction.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.