Quantification of Uncertainty with Adversarial Models

TL;DR

This paper introduces QUAM, a novel adversarial approach for more accurately quantifying epistemic uncertainty in deep learning models, outperforming existing methods especially in vision tasks.

Contribution

QUAM identifies regions with high divergence and posterior, reducing approximation error in epistemic uncertainty estimation compared to prior methods.

Findings

QUAM outperforms Deep Ensembles and MC dropout in estimating epistemic uncertainty.

QUAM effectively captures uncertainty in challenging vision tasks.

Adversarial models used in QUAM have high posterior and divergence, improving uncertainty quantification.

Abstract

Quantifying uncertainty is important for actionable predictions in real-world applications. A crucial part of predictive uncertainty quantification is the estimation of epistemic uncertainty, which is defined as an integral of the product between a divergence function and the posterior. Current methods such as Deep Ensembles or MC dropout underperform at estimating the epistemic uncertainty, since they primarily consider the posterior when sampling models. We suggest Quantification of Uncertainty with Adversarial Models (QUAM) to better estimate the epistemic uncertainty. QUAM identifies regions where the whole product under the integral is large, not just the posterior. Consequently, QUAM has lower approximation error of the epistemic uncertainty compared to previous methods. Models for which the product is large correspond to adversarial models (not adversarial examples!). Adversarial models have both a high posterior as well as a high divergence between their predictions and that of a reference model. Our experiments show that QUAM excels in capturing epistemic uncertainty for deep learning models and outperforms previous methods on challenging tasks in the vision domain.