A statistical evaluation of uncertainty disentanglement methods

Lukas Westermann

doi:10.18452/38519

August 18, 2026Open Access

A statistical evaluation of uncertainty disentanglement methods

View Full Paper

Ask AI

Bookmark

Share

Key Points

To statistically evaluate whether standard machine learning formulations accurately disentangle aleatoric and epistemic predictive uncertainty under controlled regression and classification settings.
Tested 4 data-generating processes (linear and sinusoidal, homoscedastic and heteroscedastic) across 5 controlled experimental designs: noise variation, sample scaling, extrapolation, subsampling, and omitted covariates.
Benchmarked Monte Carlo Dropout, Deep Ensembles, No-U-Turn Sampler (NUTS) estimated Bayesian neural networks, and BAMLSS under both variance- and entropy-based uncertainty decompositions.
Mutual contamination between aleatoric and epistemic uncertainty components occurred consistently across all evaluated models and decomposition formulas.
Under omitted covariate scenarios, epistemic uncertainty estimates remained unchanged while aleatoric uncertainty absorbed the model bias, misclassifying structural ignorance as data noise.
Identified that disentanglement failure stems primarily from structural and definitional limitations of existing decomposition methods rather than posterior approximation errors.

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

Simulation study finds widespread mutual contamination in predictive uncertainty disentanglement across neural methods, indicating structural limits in separating aleatoric and epistemic noise.

Abstract

No abstract yet. The journal hasn't published one we can show.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Lukas Westermann (2026) studied this question.

synapsesocial.com/papers/6a86b5318a91293e6a1cc825 https://doi.org/10.18452/38519

View Full Paper

Ask AI

Bookmark

Share

A statistical evaluation of uncertainty disentanglement methods

Key Points

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

A statistical evaluation of uncertainty disentanglement methods

Key Points

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading