In Defense of Softmax Parametrization for Calibrated and Consistent Learning to Defer

Open in new window