The work “Multiclass Local Calibration with the Jensen-Shannon Distance” with Cesare Barbera, Lorenzo Perini, Giovanni De Toni and Andrea Passerini has been accepted at AISTATS 2026!
This paper studies how local calibration and multiclass calibration are related. First, we provide a theoretical analysis of local and global multiclass calibration metrics. Then, we propose a novel method for neural network based on distilling kernel density estimators, which are provably locally calibrated. Finally, we showcase the effectiveness of our approach in an extensive experimental evaluation.
A final thanks to all the amazing co-authors involved in this project.
See you in Tangier ⛱️🌊!