Optimal Transport Dropout for Structured Predictive Uncertainty

DOI10.48550/arXiv.2609.33377MaRDI QIDQ7367485arXivFDO

Authors Giacomo Lorenzon, Francesco Regazzoni

Publication date 27 September 2026

Published in arXiv

Tldr We introduce Optimal Transport Dropout (OTD), which jointly learns the predictive mapping and law of latent perturbations, thereby capturing multimodal distributions and improving calibration and accuracy on regression tasks compared to Monte Carlo dropout and other baselines.

Full work available at URL https://doi.org/10.48550/arXiv.2609.33377

arXiv classification

cs.LGmath.NA

Cites work

• A user's guide to optimal transport
• Combining physics-based and data-driven models: advancing the frontiers of research with scientific machine learning
• Strictly Proper Scoring Rules, Prediction, and Estimation
• The Energy of Data and Distance Correlation

This page was built for publication: Optimal Transport Dropout for Structured Predictive Uncertainty

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7367485)