MAUQ-CLIP
Missingness-aware uncertainty quantification for clinical LLMs
Lab: Software Systems Lab, Montclair State University · PI: Dr. Vaibhav Anu
MAUQ-CLIP is a black-box uncertainty quantification framework for clinical large language model predictions. It is the first to treat missing evidence in a patient record as an uncertainty signal. On SEER uterine cancer records it raised accuracy from 94.0% to 98.5% and cut calibration error by 88%.
Published at IEEE HealthCom 2026. See publications.