Meta-Cal: Well-controlled Post-hoc Calibration by Ranking

                      Xingchen Ma 1 Matthew B. Blaschko 1

             Abstract               To achieve this objective, one can design models with in-
    In many applications, it is desirable that a classi   trinsically low calibration errors. These models (Wilson
    er not only makes accurate predictions, but also    et al., 2016; Pereyra et al., 2017; Lakshminarayanan et al.,
    outputs calibrated posterior probabilities. How   ...