Deterministic Uncertainty

Deterministic uncertainty methods (DUMs) aim to efficiently estimate uncertainty in deep learning models without the computational cost of traditional Bayesian approaches. Current research focuses on improving DUM calibration and accuracy through architectural innovations, such as incorporating prototype-based representations and decoupling training schemes for the core model and uncertainty estimation components. These advancements address limitations like feature collapse and improve performance on various tasks, including medical image analysis and autonomous driving perception. The resulting reliable uncertainty quantification is crucial for deploying trustworthy AI systems in high-stakes applications.

Papers

May 10, 2024

Improving Deep Learning Model Calibration for Cardiac Applications using Deterministic Uncertainty Networks and Uncertainty-aware Training
Tareen Dawood, Bram Ruijsink, Reza Razavi, Andrew P. King, Esther Puyol-Antón
Deep Learning Medical Application Uncertainty Aware Training Deterministic Uncertainty

February 20, 2024

Discriminant Distance-Aware Representation on Deterministic Uncertainty Quantification Methods
Jiaxin Zhang, Kamalika Das, Sricharan Kumar
Uncertainty Estimation Discriminative Response Deterministic Uncertainty Distance Representation

March 10, 2023

Training, Architecture, and Prior for Deterministic Uncertainty Methods
Bertrand Charpentier, Chenxiang Zhang, Stephan Günnemann
Training Data Uncertainty Estimation Architecture Design Efficient Uncertainty Estimation Deterministic Uncertainty

July 20, 2022

Latent Discriminant deterministic Uncertainty
Gianni Franchi, Xuanlong Yu, Andrei Bursuc, Emanuel Aldea, Severine Dubuisson, David Filliat
Predictive Uncertainty Uncertainty Prediction Deterministic Uncertainty

December 1, 2021

Decomposing Representations for Deterministic Uncertainty Estimation
Haiwen Huang, Joost van Amersfoort, Yarin Gal
Uncertainty Estimation Meaningful Representation Distribution Data High Uncertainty Value Unseen Test Distribution Deterministic Uncertainty

Deterministic Uncertainty

Papers

Improving Deep Learning Model Calibration for Cardiac Applications using Deterministic Uncertainty Networks and Uncertainty-aware Training

Discriminant Distance-Aware Representation on Deterministic Uncertainty Quantification Methods

Training, Architecture, and Prior for Deterministic Uncertainty Methods

Latent Discriminant deterministic Uncertainty

Decomposing Representations for Deterministic Uncertainty Estimation