Framework using UMAP-DBSCAN for unsupervised discovery of multi-modal Hidden Bias Subgroups (HBSs) in AI failure spaces. Implements a scalable Multi-Domain MMD Objective to mitigate latent Acquisition Bias and enhance robustness in clinical Ocular Disease Recognition (ODR).
computer-vision deep-learning representation-learning unsupervised-learning maximum-mean-discrepancy dbscan umap domain-adaptation explainable-ai out-of-distribution medical-ai bias-mitigation clinical-ai healthcare-ai robust-ai trustworthy-ai ocular-disease-recognition fair-ai latent-bias image-acquisition-bias
-
Updated
Dec 22, 2025 - Jupyter Notebook