Robust deep auc/auprc maximization: a new surrogate loss and empirical studies on medical image classification

Abstract

A computer-based automated method of performing classification includes learning a deep neural network by maximizing an area under a receiver operating characteristic curve (AUC) or precision-recall curve (AUPRC) score wherein a margin-based surrogate loss function is applied, receiving an input into a deep neural network, and processing the input to the deep neural network to generate a prediction, wherein the prediction comprises a classification of the input. The computer-based automated method may be performed by executing instructions in at least one processor, and wherein said instructions are stored on a non-transitory memory readable by the at least one processor.