One step closer to unbiased aleatoric uncertainty estimation

TL;DR

This paper introduces a new method for more accurately estimating aleatoric uncertainty in neural networks by actively de-noising data, addressing overestimation issues of previous variance attenuation techniques.

Contribution

The paper proposes a novel de-noising approach for aleatoric uncertainty estimation, improving accuracy over existing variance attenuation methods.

Findings

The new method better approximates true data uncertainty.

It significantly reduces overestimation of aleatoric uncertainty.

Experimental results validate the approach across various datasets.

Abstract

Neural networks are powerful tools in various applications, and quantifying their uncertainty is crucial for reliable decision-making. In the deep learning field, the uncertainties are usually categorized into aleatoric (data) and epistemic (model) uncertainty. In this paper, we point out that the existing popular variance attenuation method highly overestimates aleatoric uncertainty. To address this issue, we propose a new estimation method by actively de-noising the observed data. By conducting a broad range of experiments, we demonstrate that our proposed approach provides a much closer approximation to the actual data uncertainty than the standard method.