Deep Learning Model Uncertainty

Deep learning model uncertainty quantification aims to accurately estimate the reliability of predictions made by deep neural networks, crucial for trustworthy applications across diverse fields. Current research focuses on improving methods like Bayesian neural networks, Monte Carlo dropout, and deep ensembles, as well as exploring novel approaches such as conformal prediction and techniques that enhance the uncertainty awareness of single models. These advancements are vital for building robust and reliable AI systems, particularly in high-stakes domains like medicine, climate modeling, and engineering where understanding prediction uncertainty is paramount for informed decision-making. The ultimate goal is to develop practical and computationally efficient methods that provide well-calibrated uncertainty estimates, improving the trustworthiness and applicability of deep learning models.

Papers

June 1, 2023

Quantifying Deep Learning Model Uncertainty in Conformal Prediction
Hamed Karimi, Reza Samavi
Uncertainty Quantification Conformal Prediction Model Uncertainty Predictive Uncertainty Computational Uncertainty Deep Learning Model Uncertainty

October 25, 2022

Combined Data and Deep Learning Model Uncertainties: An Application to the Measurement of Solid Fuel Regression Rate
Georgios Georgalis, Kolos Retfalvi, Paul E. DesJardin, Abani Patra
Application Proficiency Raw Data Measurement System Uncertainty Propagation Pixel Wise Uncertainty Nuclear Fuel Deep Learning Model Uncertainty

June 27, 2022

Quantification of Deep Neural Network Prediction Uncertainties for VVUQ of Machine Learning Models
Mahmoud Yaseen, Xu Wu
Machine Learning Model Uncertainty Quantification Deep Learning Model Uncertainty Fission Gas Nuclear Engineering

May 1, 2022

A Simple Approach to Improve Single-Model Deep Uncertainty via Distance-Awareness
Jeremiah Zhe Liu, Shreyas Padhy, Jie Ren, Zi Lin, Yeming Wen, Ghassen Jerfel, Zack Nado, Jasper Snoek, Dustin Tran, Balaji Lakshminarayanan
Uncertainty Quantification Simple Approach Neural Network Gaussian Process Probabilistic Deep Learning Deep Learning Model Uncertainty

December 10, 2021

Addressing Deep Learning Model Uncertainty in Long-Range Climate Forecasting with Late Fusion
Ken C. L. Wong, Hongzhi Wang, Etienne E. Vos, Bianca Zadrozny, Campbell D. Watson, Tanveer Syeda-Mahmood
Climate Data Late Fusion Temperature Prediction Deep Learning Model Uncertainty

Deep Learning Model Uncertainty

Papers

Quantifying Deep Learning Model Uncertainty in Conformal Prediction

Combined Data and Deep Learning Model Uncertainties: An Application to the Measurement of Solid Fuel Regression Rate

Quantification of Deep Neural Network Prediction Uncertainties for VVUQ of Machine Learning Models

A Simple Approach to Improve Single-Model Deep Uncertainty via Distance-Awareness

Addressing Deep Learning Model Uncertainty in Long-Range Climate Forecasting with Late Fusion