The field of machine learning is moving towards a greater emphasis on uncertainty quantification and robustness. Recent developments have focused on improving the accuracy and reliability of models, particularly in high-stakes applications such as healthcare. Techniques such as Bayesian neural networks, variational autoencoders, and conformal prediction are being used to quantify and manage uncertainty. Additionally, there is a growing interest in epistemic artificial intelligence, which aims to develop models that can recognize and manage their own ignorance. Notable papers in this area include CoCoAFusE, which introduces a novel Bayesian covariates-dependent modeling technique, and Epistemic Wrapping, which proposes a methodology for improving uncertainty estimation in classification tasks. Overall, the field is shifting towards a more nuanced understanding of uncertainty and its role in machine learning.
Advancements in Uncertainty Quantification and Machine Learning
Sources
CoCoAFusE: Beyond Mixtures of Experts via Model Fusion
Aggregation of Dependent Expert Distributions in Multimodal Variational Autoencoders
An Approach for Handling Missing Attribute Values in Attribute-Based Access Control Policy Mining
Epistemic Wrapping for Uncertainty Quantification
Uncovering Population PK Covariates from VAE-Generated Latent Spaces
Cooperative Bayesian and variance networks disentangle aleatoric and epistemic uncertainties
Uncertainty Quantification for Machine Learning in Healthcare: A Survey
Early Prediction of Sepsis: Feature-Aligned Transfer Learning
Prediction Models That Learn to Avoid Missing Values
Learning Survival Distributions with the Asymmetric Laplace Distribution
False Promises in Medical Imaging AI? Assessing Validity of Outperformance Claims
Conformal Prediction with Corrupted Labels: Uncertain Imputation and Robust Re-weighting
Prediction via Shapley Value Regression
Clustering with Communication: A Variational Framework for Single Cell Representation Learning
Position: Epistemic Artificial Intelligence is Essential for Machine Learning Models to Know When They Do Not Know
Performance Estimation in Binary Classification Using Calibrated Confidence
Nearly Optimal Sample Complexity for Learning with Label Proportions