Title:Unreliable Uncertainty Estimates with Monte Carlo Dropout
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 neuron dropout at inference time \citep{gal2016dropout}. Hence, the method generates multiple sub-models yielding a distribution of predictions to estimate uncertainty. We empirically investigate its ability to capture true uncertainty and compare t…
Title:Unreliable Uncertainty Estimates with Monte Carlo Dropout
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 neuron dropout at inference time \citep{gal2016dropout}. Hence, the method generates multiple sub-models yielding a distribution of predictions to estimate uncertainty. We empirically 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 as observed in Bayesian models. The findings suggest that uncertainty estimates from MCD, as implemented and evaluated in these experiments, is not as reliable as those from traditional Bayesian approaches for capturing epistemic and aleatoric uncertainty.
| Comments: | Accepted for the Northern Lights Deep Learning 2026 |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2512.14851 [cs.LG] |
| (or arXiv:2512.14851v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2512.14851 arXiv-issued DOI via DataCite |
Submission history
From: Signe Riemer-Sorensen [view email] [v1] Tue, 16 Dec 2025 19:14:57 UTC (7,662 KB)