On the Learning Property of Logistic and Softmax Losses for Deep Neural Networks

Open in new window