Subjective Risk Decomposition: A New View for Uncertainty Quantification
Not on the map. Why?
Abstract (arXiv)
We present a novel viewpoint for uncertainty quantification. Uncertainty measures are not primitives, in need of axioms and argumentation, but instead consequences, of higher-level modelling decisions. We show how epistemic and aleatoric uncertainty measures can be derived via decomposition of a subjective risk, based on a strictly proper loss. Reverse cross entropy provides a prominent example, where decomposition recovers the classic information-theoretic uncertainty terms. The same approach recovers numerous measures previously proposed across the UQ literature, providing them a common theoretical foundation. This suggests a new approach to UQ: given a modelling scenario and strictly proper loss, the corresponding epistemic and aleatoric terms are induced by the subjective-risk decomposition. We then extend our view to learning theory: we introduce and analyse subjective risk analogues of excess risk, approximation error and estimation error, and identify the connections to UQ. We consider this a first step towards a full learning-theoretic framework for uncertainty quantification.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 26 of its 51 references that are in the map. First posted on arXiv 16 Jul 2026.
Builds on
26 of 51 references.
References · 26
- 2016Simple and Scalable Predictive Uncertainty Estimation using Deep Ensemblescited by 3,926
- 2017What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?cited by 2,352
- 2019Fine-Tuning Language Models from Human Preferencescited by 2,150
- 2023On-Policy Distillation of Language Models: Learning from Self-Generated Mistakescited by 939
- 2023Iterative Preference Learning from Human Feedback: Bridging Theory and Practice for RLHF under KL-Constraintcited by 350
- 2022Quantifying Aleatoric and Epistemic Uncertainty in Machine Learning: Are Conditional Entropy and Mutual Information Appropriate Measures?cited by 125
- 2019Reverse KL-Divergence Training of Prior Networks: Improved Uncertainty and Adversarial Robustnesscited by 104
- 2021DEUP: Direct Epistemic Uncertainty Predictioncited by 104
- 2022A Deeper Look into Aleatoric and Epistemic Uncertainty Disentanglementcited by 87
- 2023A Unified Theory of Diversity in Ensemble Learningcited by 71
- 2020Aleatoric and Epistemic Uncertainty with Random Forestscited by 43
- 2024Rethinking Aleatoric and Epistemic Uncertaintycited by 42
14 more
- 2022Nonparametric Uncertainty Quantification for Single Deterministic Neural Networkcited by 36
- 2023Introducing an Improved Information-Theoretic Measure of Predictive Uncertaintycited by 28
- 2024From Risk to Uncertainty: Generating Predictive Uncertainty Measures via Bayesian Estimationcited by 28
- 2024Label-wise Aleatoric and Epistemic Uncertainty Quantificationcited by 22
- 2024Quantifying Aleatoric and Epistemic Uncertainty with Proper Scoring Rulescited by 21
- 2022Uncertainty Estimates of Predictions via a General Bias-Variance Decompositioncited by 20
- 2023Second-Order Uncertainty Quantification: Variance-Based Measurescited by 19
- 2021Quantifying Epistemic Uncertainty in Deep Learningcited by 10
- 2023Efficient Uncertainty Quantification and Reduction for Over-Parameterized Neural Networkscited by 10
- 2025An Axiomatic Assessment of Entropy- and Variance-based Uncertainty Quantification in Regressioncited by 9
- 2025Position: Epistemic uncertainty estimation methods are fundamentally incompletecited by 9
- 2025A Generalized Bias-Variance Decomposition for Bregman Divergencescited by 8
- 2025Bias-variance decompositions: the exclusive privilege of Bregman divergencescited by 6
- 2025Uncertainty Quantification for Regression using Proper Scoring Rulescited by 5
Built on by
Citing papers per year: 2026: 1.
1 paper.