On Calibration of Modern Neural Networks

Chuan Guo, Geoff Pleiss, Yu Sun +1

Confidence calibration -- the problem of predicting probability estimates representative of the true correctness likelihood -- is important for classification models in many applic…