Pathological Regularization Regimes in Classification Tasks

Maximilian Wiesmann, Paul Larsen

In this paper we demonstrate the possibility of a trend reversal in binary classification tasks between the dataset and a classification score obtained from a trained model. This t…