Reproduce “MedHEB: Benchmarking Medical Embeddings Across Heterogeneous Clinical Evidence” 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
Yingshu Li, Shaoyang Zhou, Zhanyu Wang, Yunyi Liu, Xinyu Liang, Xi Zhang, Lingqiao Liu, Lei Wang and 1 more
MedHEB tests clinical evidence retrieval across 107 tasks from 54 datasets, using a shared ranking protocol across image, modality-bridging, text, question-conditioned, and region-level relations. It reports MedEmb-8B, fine-tuned on in-distribution training data, achieves 49.32% Hit@1, versus 24.64% for Qwen3-VL-Embed-8B. This is a benchmark result; on 11 held-out-source tasks, MedEmb-8B scores 44.22%, below MedEmb-2B’s 45.77%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.