Analysis of Softmax Approximation for Deep Classifiers under Input-Dependent Label Noise

Open in new window