GEM-FI: Gated Evidential Mixtures with Fisher Modulation

2026-05-05

AI Summary

This work addresses the persistent challenges in evidential deep learning—namely, overconfidence, poor calibration, and the inability to model multimodal epistemic uncertainty within a single forward pass. To this end, we propose the GEM family of models, which, for the first time, enables multimodal epistemic uncertainty estimation in a single inference step. Our approach employs energy-based gating of evidential outputs, a lightweight mixture-of-evidence head, and a learnable routing mechanism, further stabilized by Fisher information regularization to prevent head collapse, ensure smooth boundary uncertainty, and regularize mixture assignments. Experiments demonstrate that GEM achieves a 2.64% accuracy gain on CIFAR-10, reduces the Brier score by 7.46, attains a 99.94% AUPR in misclassification detection, and significantly outperforms existing baselines in out-of-distribution detection.

Abstract

Evidential Deep Learning (EDL) enables single-pass uncertainty estimation by predicting Dirichlet evidence, but it can remain overconfident and poorly calibrated, and it often fails to represent multi-modal epistemic uncertainty. We introduce Gated Evidential Mixtures (GEM), a family of models that learns an in-model energy signal and uses it to gate evidential outputs end-to-end in a distance-informed manner. GEM-CORE learns a feature-level energy and maps it to a bounded gate that smoothly suppresses evidence when support is low. To capture epistemic multi-modality without multi-pass ensembling, GEM-MIX adds a lightweight mixture of evidential heads with learned routing weights while preserving single-pass inference. Finally, GEM-FI stabilizes mixture allocations via a Fisher-informed regularizer, reducing head collapse and producing smoother boundary uncertainty. Across image classification and OOD detection benchmarks, GEM improves calibration and ID/OOD separation with single-pass inference. On CIFAR-10, GEM-FI vs. DAEDL improves accuracy from 91.11 to 93.75 (+2.64 pp), reduces Brier x100 from 14.27 to 6.81 (-7.46), and also improves misclassification-detection AUPR from 99.08 to 99.94 (+0.86). For epistemic OOD detection, GEM-FI achieves AUPR/AUROC of 92.59/95.09 on CIFAR-10 to SVHN and 90.20/89.06 on CIFAR-10 to CIFAR-100, compared with 85.54/89.30 and 88.19/86.10 for DAEDL.

Problem

• Evidential Deep Learning
• Uncertainty Estimation
• Epistemic Uncertainty
• Model Calibration
• Out-of-Distribution Detection

Innovation

• Evidential Deep Learning
• Uncertainty Estimation
• Single-pass Inference
• Mixture of Experts
• Fisher Information Regularization