Reproduce “SMI: Ontology-Grounded Semantic IDs for Hierarchy-Aware Medical Concept Representation” 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

Ziyang Song, Lia Shen, Yixuan Li, Qincheng Lu, Weihao Li, Haoran Xu, He Zhu, Xinyu Wang and 1 more

SMI represents medical concepts by summing frozen biomedical language-model embeddings along expert-defined code paths. SMI-W adds validation-selected weights without training an ontology encoder. On MIMIC-IV, SMI-W reaches 44.2% Recall@15 for 98 rare ATC codes, versus 40.7% for G-BERT. This retrospective benchmark result is not evidence of clinical outcomes or training-free downstream prediction.

Summary by alphaXiv.

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