Uncertainty Disentanglement

Uncertainty disentanglement in machine learning aims to separate the inherent randomness in data (aleatoric uncertainty) from the uncertainty due to model limitations (epistemic uncertainty), providing a more nuanced understanding of prediction reliability. Current research focuses on developing and benchmarking methods for achieving this separation, often employing techniques like ensembles, dropout, and Gaussian processes, and evaluating their performance across various tasks including out-of-distribution detection and active learning. While progress has been made, studies consistently reveal that perfectly disentangling these uncertainty sources remains a significant challenge, highlighting the need for improved methods and a deeper understanding of their interactions. This work is crucial for building more trustworthy and reliable AI systems, particularly in high-stakes applications like medicine.

Papers

November 3, 2024

Know Where You're Uncertain When Planning with Multimodal Foundation Models: A Formal Framework
Neel P. Bhatt, Yunhao Yang, Rohan Siva, Daniel Milan, Ufuk Topcu, Zhangyang Wang
Full Model Multimodal Foundation Model Complex Scene Disentanglement Framework Perceptual Uncertainty Uncertainty Disentanglement

August 22, 2024

How disentangled are your classification uncertainties?
Ivo Pascal de Jong, Andreea Ioana Sburlea, Matias Valdenegro-Toro
Uncertainty Quantification Epistemic Uncertainty Disentanglement Capability Disentangle Content Classification Uncertainty Uncertainty Disentanglement

February 29, 2024

Benchmarking Uncertainty Disentanglement: Specialized Uncertainties for Specialized Tasks
Bálint Mucsányi, Michael Kirchhof, Seong Joon Oh
Uncertainty Quantification Uncertainty Estimation Disentanglement Capability Aleatoric Uncertainty High Uncertainty Value Task Specification Quantified Uncertainty Uncertainty Disentanglement

November 11, 2022

Disentangled Uncertainty and Out of Distribution Detection in Medical Generative Models
Kumud Lakara, Matias Valdenegro-Toro
Uncertainty Quantification Distribution Detection Image to Image Translation Epistemic Uncertainty High Uncertainty Value Medical Generative Uncertainty Disentanglement

October 20, 2022

Uncertainty Disentanglement with Non-stationary Heteroscedastic Gaussian Processes for Active Learning
Zeel B Patel, Nipun Batra, Kevin Murphy
Active Learning Gaussian Process Epistemic Uncertainty Bayesian Nonparametric Uncertainty Disentanglement

April 20, 2022

A Deeper Look into Aleatoric and Epistemic Uncertainty Disentanglement
Matias Valdenegro-Toro, Daniel Saromo
Uncertainty Quantification Epistemic Uncertainty Aleatoric Uncertainty Closer Look Non Gaussian Aleatoric Uncertainty Uncertainty Disentanglement

Uncertainty Disentanglement

Papers

Know Where You're Uncertain When Planning with Multimodal Foundation Models: A Formal Framework

How disentangled are your classification uncertainties?

Benchmarking Uncertainty Disentanglement: Specialized Uncertainties for Specialized Tasks

Disentangled Uncertainty and Out of Distribution Detection in Medical Generative Models

Uncertainty Disentanglement with Non-stationary Heteroscedastic Gaussian Processes for Active Learning

A Deeper Look into Aleatoric and Epistemic Uncertainty Disentanglement