Uncertainty Decomposition

Uncertainty decomposition aims to separate the total uncertainty of a prediction into its constituent parts, primarily aleatoric (inherent data randomness) and epistemic (model limitations) uncertainty. Current research focuses on applying this decomposition to various models, including Bayesian neural networks, large language models, and deep neural networks, often using techniques like ensembling and input clarification to quantify these uncertainty types. This work is crucial for improving the reliability and trustworthiness of machine learning models across diverse fields, from medical diagnosis and autonomous driving to scientific machine learning, by providing insights into the sources of prediction errors and enabling more informed decision-making.

Papers

September 17, 2024

Uncertainty Decomposition and Error Margin Detection of Homodyned-K Distribution in Quantitative Ultrasound
Dorsa Ameri, Ali K. Z. Tehrani, Ivan M. Rosado-Mendez, Hassan Rivaz
Bayesian Neural Network Bayesian Model Quantitative Ultrasound Error Simulation K Distribution Uncertainty Decomposition

June 14, 2024

Know the Unknown: An Uncertainty-Sensitive Method for LLM Instruction Tuning
Jiaqi Li, Yixuan Tang, Yi Yang
Large Language Model Instruction Tuning Targeted Activation Penalty Robust Fine Tuning Uncertainty Decomposition

November 15, 2023

Decomposing Uncertainty for Large Language Models through Input Clarification Ensembling
Bairu Hou, Yujian Liu, Kaizhi Qian, Jacob Andreas, Shiyu Chang, Yang Zhang
Large Language Model High Uncertainty Anticipation Aleatoric Uncertainty Uncertainty Decomposition

April 20, 2023

Multidimensional Uncertainty Quantification for Deep Neural Networks
Xujiang Zhao
Deep Neural Network High Uncertainty Value Probabilistic Fusion Uncertainty Decomposition High Dimensional Uncertainty

November 17, 2022

Introduction and Exemplars of Uncertainty Decomposition
Shuo Chen
Gaussian Process High Uncertainty Anticipation Ensemble Learning Gentle Introduction Exemplar Image Uncertainty Decomposition

September 5, 2022

A Robust Learning Methodology for Uncertainty-aware Scientific Machine Learning models
Erbet Costa Almeida, Carine de Menezes Rebello, Marcio Fontana, Leizer Schnitman, Idelfonso Bessa dos Reis Nogueira
Uncertainty Estimation Robust Learning Uncertainty Aware Model Uncertainty Decomposition Robust Learning Method SciML Model

July 11, 2022

What is Flagged in Uncertainty Quantification? Latent Density Models for Uncertainty Categorization
Hao Sun, Boris van Breugel, Jonathan Crabbe, Nabeel Seedat, Mihaela van der Schaar
Uncertainty Quantification Confusion Matrix Uncertainty Decomposition Misclassification Detection

March 2, 2022

MUAD: Multiple Uncertainties for Autonomous Driving, a benchmark for multiple uncertainty types and tasks
Gianni Franchi, Xuanlong Yu, Andrei Bursuc, Angel Tena, Rémi Kazmierczak, Séverine Dubuisson, Emanuel Aldea, David Filliat
Autonomous Driving New Benchmark MAESTRO Dataset New Task Predictive Uncertainty Uncertainty Decomposition Total Uncertainty

March 1, 2022

Uncertainty categories in medical image segmentation: a study of source-related diversity
Luke Whitbread, Mark Jenkinson
Deep Learning Medical Image Segmentation Study Feature High Uncertainty Value BRAT Challenge Uncertainty Decomposition

Uncertainty Decomposition

Papers

Uncertainty Decomposition and Error Margin Detection of Homodyned-K Distribution in Quantitative Ultrasound

Know the Unknown: An Uncertainty-Sensitive Method for LLM Instruction Tuning

Decomposing Uncertainty for Large Language Models through Input Clarification Ensembling

Multidimensional Uncertainty Quantification for Deep Neural Networks

Introduction and Exemplars of Uncertainty Decomposition

A Robust Learning Methodology for Uncertainty-aware Scientific Machine Learning models

What is Flagged in Uncertainty Quantification? Latent Density Models for Uncertainty Categorization

MUAD: Multiple Uncertainties for Autonomous Driving, a benchmark for multiple uncertainty types and tasks

Uncertainty categories in medical image segmentation: a study of source-related diversity