Unreliable Monte Carlo Dropout Uncertainty Estimation

Aslak Djupskås, Signe Riemer-Sørensen, Alexander Johannes Stasik
Proceedings of the 7th Northern Lights Deep Learning Conference (NLDL), PMLR 307:106-114, 2026.

Abstract

Reliable uncertainty estimation is crucial for machine learning models, especially in safety-critical domains. While exact Bayesian inference offers a principled approach, it is often computationally infeasible for deep neural networks. Monte Carlo dropout (MCD) was proposed as an efficient approximation to Bayesian inference in deep learning by applying dropout at inference time. Hence, the method generates multiple sub-models yielding a distribution of predictions to estimate uncertainty. We investigate its ability to capture true uncertainty and compare to Gaussian Processes (GP) and Bayesian Neural Networks (BNN). We find that MCD struggles to accurately reflect the underlying true uncertainty, particularly failing to capture increased uncertainty in extrapolation and interpolation regions observed in Bayesian models. The findings suggest that uncertainty estimates from MCD, as implemented and evaluated in these experiments, may not be as reliable as those from traditional Bayesian approaches for capturing epistemic and aleatoric uncertainty.

Cite this Paper

BibTeX
@InProceedings{pmlr-v307-djupskas26a, title = {Unreliable Monte Carlo Dropout Uncertainty Estimation}, author = {Djupsk{\r{a}}s, Aslak and Riemer-S{\o}rensen, Signe and Stasik, Alexander Johannes}, booktitle = {Proceedings of the 7th Northern Lights Deep Learning Conference (NLDL)}, pages = {106--114}, year = {2026}, editor = {Kim, Hyeongji and Ramírez Rivera, Adín and Ricaud, Benjamin}, volume = {307}, series = {Proceedings of Machine Learning Research}, month = {06--08 Jan}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v307/main/assets/djupskas26a/djupskas26a.pdf}, url = {https://proceedings.mlr.press/v307/djupskas26a.html}, abstract = {Reliable uncertainty estimation is crucial for machine learning models, especially in safety-critical domains. While exact Bayesian inference offers a principled approach, it is often computationally infeasible for deep neural networks. Monte Carlo dropout (MCD) was proposed as an efficient approximation to Bayesian inference in deep learning by applying dropout at inference time. Hence, the method generates multiple sub-models yielding a distribution of predictions to estimate uncertainty. We investigate its ability to capture true uncertainty and compare to Gaussian Processes (GP) and Bayesian Neural Networks (BNN). We find that MCD struggles to accurately reflect the underlying true uncertainty, particularly failing to capture increased uncertainty in extrapolation and interpolation regions observed in Bayesian models. The findings suggest that uncertainty estimates from MCD, as implemented and evaluated in these experiments, may not be as reliable as those from traditional Bayesian approaches for capturing epistemic and aleatoric uncertainty.} }
Endnote
%0 Conference Paper %T Unreliable Monte Carlo Dropout Uncertainty Estimation %A Aslak Djupskås %A Signe Riemer-Sørensen %A Alexander Johannes Stasik %B Proceedings of the 7th Northern Lights Deep Learning Conference (NLDL) %C Proceedings of Machine Learning Research %D 2026 %E Hyeongji Kim %E Adín Ramírez Rivera %E Benjamin Ricaud %F pmlr-v307-djupskas26a %I PMLR %P 106--114 %U https://proceedings.mlr.press/v307/djupskas26a.html %V 307 %X Reliable uncertainty estimation is crucial for machine learning models, especially in safety-critical domains. While exact Bayesian inference offers a principled approach, it is often computationally infeasible for deep neural networks. Monte Carlo dropout (MCD) was proposed as an efficient approximation to Bayesian inference in deep learning by applying dropout at inference time. Hence, the method generates multiple sub-models yielding a distribution of predictions to estimate uncertainty. We investigate its ability to capture true uncertainty and compare to Gaussian Processes (GP) and Bayesian Neural Networks (BNN). We find that MCD struggles to accurately reflect the underlying true uncertainty, particularly failing to capture increased uncertainty in extrapolation and interpolation regions observed in Bayesian models. The findings suggest that uncertainty estimates from MCD, as implemented and evaluated in these experiments, may not be as reliable as those from traditional Bayesian approaches for capturing epistemic and aleatoric uncertainty.
APA
Djupskås, A., Riemer-Sørensen, S. & Stasik, A.J.. (2026). Unreliable Monte Carlo Dropout Uncertainty Estimation. Proceedings of the 7th Northern Lights Deep Learning Conference (NLDL), in Proceedings of Machine Learning Research 307:106-114 Available from https://proceedings.mlr.press/v307/djupskas26a.html.

Related Material