On the Calibration of Epistemic Uncertainty: Principles, Paradoxes and Conflictual Loss

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Fellaji, Mohammed, Pennerath, Frédéric, Conan-Guez, Brieuc, Couceiro, Miguel
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910530860482560
author Fellaji, Mohammed
Pennerath, Frédéric
Conan-Guez, Brieuc
Couceiro, Miguel
author_facet Fellaji, Mohammed
Pennerath, Frédéric
Conan-Guez, Brieuc
Couceiro, Miguel
contents The calibration of predictive distributions has been widely studied in deep learning, but the same cannot be said about the more specific epistemic uncertainty as produced by Deep Ensembles, Bayesian Deep Networks, or Evidential Deep Networks. Although measurable, this form of uncertainty is difficult to calibrate on an objective basis as it depends on the prior for which a variety of choices exist. Nevertheless, epistemic uncertainty must in all cases satisfy two formal requirements: first, it must decrease when the training dataset gets larger and, second, it must increase when the model expressiveness grows. Despite these expectations, our experimental study shows that on several reference datasets and models, measures of epistemic uncertainty violate these requirements, sometimes presenting trends completely opposite to those expected. These paradoxes between expectation and reality raise the question of the true utility of epistemic uncertainty as estimated by these models. A formal argument suggests that this disagreement is due to a poor approximation of the posterior distribution rather than to a flaw in the measure itself. Based on this observation, we propose a regularization function for deep ensembles, called conflictual loss in line with the above requirements. We emphasize its strengths by showing experimentally that it restores both requirements of epistemic uncertainty, without sacrificing either the performance or the calibration of the deep ensembles.
format Preprint
id arxiv_https___arxiv_org_abs_2407_12211
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On the Calibration of Epistemic Uncertainty: Principles, Paradoxes and Conflictual Loss
Fellaji, Mohammed
Pennerath, Frédéric
Conan-Guez, Brieuc
Couceiro, Miguel
Machine Learning
The calibration of predictive distributions has been widely studied in deep learning, but the same cannot be said about the more specific epistemic uncertainty as produced by Deep Ensembles, Bayesian Deep Networks, or Evidential Deep Networks. Although measurable, this form of uncertainty is difficult to calibrate on an objective basis as it depends on the prior for which a variety of choices exist. Nevertheless, epistemic uncertainty must in all cases satisfy two formal requirements: first, it must decrease when the training dataset gets larger and, second, it must increase when the model expressiveness grows. Despite these expectations, our experimental study shows that on several reference datasets and models, measures of epistemic uncertainty violate these requirements, sometimes presenting trends completely opposite to those expected. These paradoxes between expectation and reality raise the question of the true utility of epistemic uncertainty as estimated by these models. A formal argument suggests that this disagreement is due to a poor approximation of the posterior distribution rather than to a flaw in the measure itself. Based on this observation, we propose a regularization function for deep ensembles, called conflictual loss in line with the above requirements. We emphasize its strengths by showing experimentally that it restores both requirements of epistemic uncertainty, without sacrificing either the performance or the calibration of the deep ensembles.
title On the Calibration of Epistemic Uncertainty: Principles, Paradoxes and Conflictual Loss
topic Machine Learning
url https://arxiv.org/abs/2407.12211