EviNAM: Intelligibility and Uncertainty via Evidential Neural Additive Models

TL;DR

EviNAM addresses the need for interpretable and trustworthy predictions by uniting the additive interpretability of Neural Additive Models with evidential uncertainty estimation from DER. It introduces a per-feature, additive distributional parameterization under a Normal-Inverse-Gamma prior, enabling explicit contributions and simultaneous estimation of aleatoric and epistemic uncertainty in a single forward pass. The method preserves additivity through feature-level forwarding of nonlinearities and extends to classification via Dirichlet-based evidence while maintaining per-feature contributions. Empirical results show competitive predictive performance on real data and clear intelligibility benefits, along with practical implications for detecting out-of-domain inputs and communicating model confidence.

Abstract

Intelligibility and accurate uncertainty estimation are crucial for reliable decision-making. In this paper, we propose EviNAM, an extension of evidential learning that integrates the interpretability of Neural Additive Models (NAMs) with principled uncertainty estimation. Unlike standard Bayesian neural networks and previous evidential methods, EviNAM enables, in a single pass, both the estimation of the aleatoric and epistemic uncertainty as well as explicit feature contributions. Experiments on synthetic and real data demonstrate that EviNAM matches state-of-the-art predictive performance. While we focus on regression, our method extends naturally to classification and generalized additive models, offering a path toward more intelligible and trustworthy predictions.

Figures (8)

• Figure : (a) $\text{DER}_{\text{MLP}}$
• Figure : (a) Feature median income
• Figure : (a) $\text{DER}_{\text{MLP}}$
• Figure : (b) EviNAM
• Figure : (c) $\text{DER}_{\text{MLP}}$