AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

From Knowledge to Inference: Formalizing Specialized Public Health Reasoning on GlobalHealthAtlas

Zhaokun Yan, Shan Xu, Wuzheng Dong, Zhaohan Liu, Lijie Feng, Chengxiao Dai, Chen Tianqi, Binfan Liu, Yunpu Ma, Wenting Wei, Yingting Li, Yi Zhang and 1 more

ICML 2026 · Large language models › Retrieval and knowledge · approach: Domain-specific QA datasets

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add SFT on GlobalHealthAtlas (ablation)qwen3-4b · GlobalHealthAtlasscore: 5.773 -> 6.382Improved
add SFT on GlobalHealthAtlas (ablation)qwen3.5-122B-A10B · Instruction-public-health-datasetscore: 4.199 -> 6.669Improved
add SFT on GlobalHealthAtlas (ablation)qwen3-8b · GlobalHealthAtlasscore: 6.490 -> 6.532No clear effect
add SFT on GlobalHealthAtlas (ablation)qwen3-8b · publichealth-qascore: 6.075 -> 5.652Hurt
add SFT on GlobalHealthAtlas (ablation)Llama-3.2-3B-Instruct · MedQAscore: 6.326 -> 6.235Hurt
add SFT on GlobalHealthAtlas (ablation)Llama-3.3-70B-Instruct · MedQAscore: 6.266 -> 7.937Improved
increase SFT data from 60% to 100% (ablation)qwen3-14b · GlobalHealthAtlasscore: 6.5748 -> 6.5648No clear effect
add SFT data 10% (vs none) (ablation)qwen3-8b · GlobalHealthAtlasscore: 6.4896 -> 6.4185Hurt
add SFT on GlobalHealthAtlas (ablation)qwen3-14b · GPQAscore: 5.277 -> 5.723Improved
add SFT on GlobalHealthAtlas (ablation)qwen3-14b · MMLU-proscore: 7.426 -> 7.309Hurt
add 0.85 weight interpolation between SFT and base (ablation)qwen3-14b SFT · MMLU-proscore: 7.309 -> 7.545Improved
use uniform instead of expert quality weights (ablation)Quality gate scoring · Weighting sensitivity (SC format)Kendall tau: 0.930 -> 0.775Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy