Why Machine Learning Models Fail to Fully Capture Epistemic Uncertainty

2025-05-29

AI Summary

Existing second-order uncertainty modeling methods systematically underestimate epistemic uncertainty, primarily because they neglect model bias—a critical source of epistemic uncertainty—and erroneously attribute its error to aleatoric uncertainty. Method: This work explicitly incorporates model bias into the epistemic uncertainty taxonomy, proposing a fine-grained classification framework and an extended bias–variance decomposition theory. We further design a synthetic-data-based evaluation protocol to simulate multi-source epistemic uncertainty. Contribution/Results: We theoretically prove that high model bias leads to substantial underestimation of epistemic uncertainty. Empirical results show that mainstream methods underestimate epistemic uncertainty by 42%–68% on average; moreover, aleatoric uncertainty estimates are reliable only when model bias is well-controlled. This work establishes a more rigorous foundation for epistemic uncertainty modeling in trustworthy machine learning.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Multiagent Systems: Multiagent Systems under Uncertainty

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics

Abstract

In recent years various supervised learning methods that disentangle aleatoric and epistemic uncertainty based on second-order distributions have been proposed. We argue that these methods fail to capture critical components of epistemic uncertainty, particularly due to the often-neglected component of model bias. To show this, we make use of a more fine-grained taxonomy of epistemic uncertainty sources in machine learning models, and analyse how the classical bias-variance decomposition of the expected prediction error can be decomposed into different parts reflecting these uncertainties. By using a simulation-based evaluation protocol which encompasses epistemic uncertainty due to both procedural- and data-driven uncertainty components, we illustrate that current methods rarely capture the full spectrum of epistemic uncertainty. Through theoretical insights and synthetic experiments, we show that high model bias can lead to misleadingly low estimates of epistemic uncertainty, and common second-order uncertainty quantification methods systematically blur bias-induced errors into aleatoric estimates, thereby underrepresenting epistemic uncertainty. Our findings underscore that meaningful aleatoric estimates are feasible only if all relevant sources of epistemic uncertainty are properly represented.

Problem

• Current methods fail to capture full epistemic uncertainty spectrum
• Model bias leads to misleadingly low epistemic uncertainty estimates
• Second-order methods blur bias-induced errors into aleatoric estimates

Innovation

• Second-order distributions disentangle uncertainty types
• Simulation protocol evaluates epistemic uncertainty sources
• Bias-variance decomposition reveals uncertainty underestimation