Credal Ensemble Distillation for Uncertainty Quantification

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Kaizheng, Cuzzolin, Fabio, Moens, David, Hallez, Hans
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912715409195008
author Wang, Kaizheng
Cuzzolin, Fabio
Moens, David
Hallez, Hans
author_facet Wang, Kaizheng
Cuzzolin, Fabio
Moens, David
Hallez, Hans
contents Deep ensembles (DE) have emerged as a powerful approach for quantifying predictive uncertainty and distinguishing its aleatoric and epistemic components, thereby enhancing model robustness and reliability. However, their high computational and memory costs during inference pose significant challenges for wide practical deployment. To overcome this issue, we propose credal ensemble distillation (CED), a novel framework that compresses a DE into a single model, CREDIT, for classification tasks. Instead of a single softmax probability distribution, CREDIT predicts class-wise probability intervals that define a credal set, a convex set of probability distributions, for uncertainty quantification. Empirical results on out-of-distribution detection benchmarks demonstrate that CED achieves superior or comparable uncertainty estimation compared to several existing baselines, while substantially reducing inference overhead compared to DE.
format Preprint
id arxiv_https___arxiv_org_abs_2511_13766
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Credal Ensemble Distillation for Uncertainty Quantification
Wang, Kaizheng
Cuzzolin, Fabio
Moens, David
Hallez, Hans
Machine Learning
Artificial Intelligence
Deep ensembles (DE) have emerged as a powerful approach for quantifying predictive uncertainty and distinguishing its aleatoric and epistemic components, thereby enhancing model robustness and reliability. However, their high computational and memory costs during inference pose significant challenges for wide practical deployment. To overcome this issue, we propose credal ensemble distillation (CED), a novel framework that compresses a DE into a single model, CREDIT, for classification tasks. Instead of a single softmax probability distribution, CREDIT predicts class-wise probability intervals that define a credal set, a convex set of probability distributions, for uncertainty quantification. Empirical results on out-of-distribution detection benchmarks demonstrate that CED achieves superior or comparable uncertainty estimation compared to several existing baselines, while substantially reducing inference overhead compared to DE.
title Credal Ensemble Distillation for Uncertainty Quantification
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2511.13766