![[2602.21160] Not Just How Much, But Where: Decomposing Epistemic Uncertainty into Per-Class Contributions](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.21160] Not Just How Much, But Where: Decomposing Epistemic Uncertainty into Per-Class Contributions
Summary
The paper presents a novel method for decomposing epistemic uncertainty in machine learning models into per-class contributions, enhancing safety in classification tasks.
Why It Matters
Understanding and managing epistemic uncertainty is crucial in safety-critical applications. This research provides a framework that allows for better decision-making by distinguishing uncertainty across different classes, which can significantly reduce risks in areas like healthcare and autonomous systems.
Key Takeaways
- Introduces a per-class vector for decomposing mutual information, enhancing uncertainty measurement.
- Demonstrates improved performance in selective prediction and out-of-distribution detection tasks.
- Highlights the importance of how uncertainty is propagated through neural networks.
Statistics > Machine Learning arXiv:2602.21160 (stat) [Submitted on 24 Feb 2026] Title:Not Just How Much, But Where: Decomposing Epistemic Uncertainty into Per-Class Contributions Authors:Mame Diarra Toure, David A. Stephens View a PDF of the paper titled Not Just How Much, But Where: Decomposing Epistemic Uncertainty into Per-Class Contributions, by Mame Diarra Toure and 1 other authors View PDF HTML (experimental) 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-critical class. We decompose MI into a per-class vector $C_k(x)=\sigma_k^{2}/(2\mu_k)$, with $\mu_k{=}\mathbb{E}[p_k]$ and $\sigma_k^2{=}\mathrm{Var}[p_k]$ across posterior samples. The decomposition follows from a second-order Taylor expansion of the entropy; the $1/\mu_k$ weighting corrects boundary suppression and makes $C_k$ comparable across rare and common classes. By construction $\sum_k C_k \approx \mathrm{MI}$, and a companion skewness diagnostic flags inputs where the approximation degrades. After characterising the axiomatic properties of $C_k$, we validate it on three tasks: (i) selective prediction for diabetic retinopathy, where critical-class $C_k$ reduces selective risk by 34.7\% over MI and 56.2\% over variance baselines; (ii) out-of-distribution detection on clinical and image...
