Density-Informed Pseudo-Counts for Calibrated Evidential Deep Learning

TL;DR

This paper offers a theoretical interpretation of Evidential Deep Learning, identifies its overconfidence issue on out-of-distribution data, and proposes a new density-informed parametrization to improve uncertainty calibration and robustness.

Contribution

It introduces Density-Informed Pseudo-count EDL (DIP-EDL), a novel approach that decouples class prediction from uncertainty magnitude, enhancing calibration and robustness.

Findings

DIP-EDL achieves asymptotic concentration.

Improves uncertainty calibration under distributional shift.

Enhances interpretability and robustness.

Abstract

Evidential Deep Learning (EDL) is a popular framework for uncertainty-aware classification that models predictive uncertainty via Dirichlet distributions parameterized by neural networks. Despite its popularity, its theoretical foundations and behavior under distributional shift remain poorly understood. In this work, we provide a principled statistical interpretation by proving that EDL training corresponds to amortized variational inference in a hierarchical Bayesian model with a tempered pseudo-likelihood. This perspective reveals a major drawback: standard EDL conflates epistemic and aleatoric uncertainty, leading to systematic overconfidence on out-of-distribution (OOD) inputs. To address this, we introduce Density-Informed Pseudo-count EDL (DIP-EDL), a new parametrization that decouples class prediction from the magnitude of uncertainty by separately estimating the conditional label distribution and the marginal covariate density. This separation preserves evidence in high-density regions while shrinking predictions toward a uniform prior for OOD data. Theoretically, we prove that DIP-EDL achieves asymptotic concentration. Empirically, we show that our method enhances interpretability and improves robustness and uncertainty calibration under distributional shift.