arXiv stat.ML
· Papers
Not Just How Much, But Where: Decomposing Epistemic Uncertainty into Per-Class Contributions
arXiv:2602.21160v4 Announce Type: replace Abstract: In safety-critical classification, the cost of failure is often asymmetric, yet Bayesian deep learning summarises epistemic uncertainty with a single scalar, mutual information (MI), that cannot distinguish whether a model's ignorance involves a benign or safety-criti