Abstract: The performance of discriminative machine-learning classifiers, such as neural networks, is limited by training label inconsistencies. Even expert-based annotations can suffer from label ...