Predictive Uncertainty Quantification

Predictive uncertainty quantification (PUQ) aims to estimate the reliability of machine learning model predictions, crucial for trustworthy decision-making in various applications. Current research focuses on improving PUQ methods for deep learning models, particularly exploring Bayesian neural networks, evidential deep learning, and adversarial approaches to better capture both aleatoric (data-inherent) and epistemic (model-inherent) uncertainty. These advancements are vital for enhancing the robustness and explainability of AI systems across domains like autonomous driving and natural language processing, where reliable uncertainty estimates are essential for safe and effective deployment.

Papers

May 31, 2024

Uncertainty Quantification for Bird's Eye View Semantic Segmentation: Methods and Benchmarks
Linlin Yu, Bowen Yang, Tianhao Wang, Kangshuo Li, Feng Chen
Semantic Segmentation New Benchmark Uncertainty Quantification NCD Method Bird'S Eye View Cross Entropy Eye View Bird Specie Segmentation Quality Predictive Uncertainty Quantification

April 27, 2024

Implicit Generative Prior for Bayesian Neural Networks
Yijia Liu, Xiao Wang
Bayesian Neural Network Empirical Bayes Implicit Generative Sparse Variational Predictive Uncertainty Quantification

February 16, 2024

Predictive Uncertainty Quantification via Risk Decompositions for Strictly Proper Scoring Rules
Nikita Kotelevskii, Maxim Panov
Uncertainty Quantification Epistemic Uncertainty Bayesian Approach Scoring Rule Hierarchical Risk Parity Bayes Risk Predictive Uncertainty Quantification

February 9, 2024

Are Uncertainty Quantification Capabilities of Evidential Deep Learning a Mirage?
Maohao Shen, J. Jon Ryu, Soumya Ghosh, Yuheng Bu, Prasanna Sattigeri, Subhro Das, Gregory W. Wornell
Epistemic Uncertainty Model Uncertainty Evidential Deep Learning Predictive Uncertainty Quantification

July 6, 2023

Quantification of Uncertainty with Adversarial Models
Kajetan Schweighofer, Lukas Aichberger, Mykyta Ielanskyi, Günter Klambauer, Sepp Hochreiter
Deep Learning Model Adversarial Example High Uncertainty Anticipation Epistemic Uncertainty Adversarial Model Predictive Uncertainty Quantification

July 3, 2023

Shifting Attention to Relevance: Towards the Predictive Uncertainty Quantification of Free-Form Large Language Models
Jinhao Duan, Hao Cheng, Shiqi Wang, Alex Zavalny, Chenan Wang, Renjing Xu, Bhavya Kailkhura, Kaidi Xu
Uncertainty Quantification Question Answering Information Redundancy Unstructured Text Relative Relevance Autoregressive LLM Predictive Uncertainty Quantification Attention Transfer

November 3, 2022

Quantifying Model Uncertainty for Semantic Segmentation using Operators in the RKHS
Rishabh Singh, Jose C. Principe
Semantic Segmentation Semantic Segmentation Model Model Uncertainty Reproducing Kernel Hilbert Space Feature Operator Road Segmentation Predictive Uncertainty Quantification

Predictive Uncertainty Quantification

Papers

Uncertainty Quantification for Bird's Eye View Semantic Segmentation: Methods and Benchmarks

Implicit Generative Prior for Bayesian Neural Networks

Predictive Uncertainty Quantification via Risk Decompositions for Strictly Proper Scoring Rules

Are Uncertainty Quantification Capabilities of Evidential Deep Learning a Mirage?

Quantification of Uncertainty with Adversarial Models

Shifting Attention to Relevance: Towards the Predictive Uncertainty Quantification of Free-Form Large Language Models

Quantifying Model Uncertainty for Semantic Segmentation using Operators in the RKHS