Coby Penso

Confidence Calibration of Deep Learning Systems

This thesis develops methods for improving calibration under label noise and studies calibration in unsupervised domain adaptation, where a model trained on a labeled source domain is adapted to an unlabeled target domain.