2019 (modified: 11 Nov 2022)ICML 2019Readers: Everyone
Abstract:We introduce a novel method to combat label noise when training deep neural networks for classification. We propose a loss function that permits abstention during training thereby allowing the DNN ...