Locally Adaptive Label Smoothing for Predictive Churn

                        Dara Bahri 1 Heinrich Jiang 1

              Abstract               as we will show later in the experiments. Thus, even an
                                 optimized training procedure can lead to high prediction
    Training modern neural networks is an inherently
                                 churn, which refers to the proportion of sample-level dis-
    noisy process that can lead to high prediction
            ...