Calibration in One Go: Jointly Learning Accurate and Calibrated Neural Networks
This paper shows that standard (post-hoc) calibrators can be trained end-to-end with the backbone via a bilevel Stackelberg formulation, and shows that using the final classification layer itself as the calibrator yields accurate, well-calibrated models with no additional parameters.
Nektarios G. Kalampalikis, Pablo M. Olmos, Isabel Valera
· 0 citations