Towards Understanding Learning in Neural Networks with Linear Teachers

                  Roei Sarussi 1 Alon Brutzkus 1 Amir Globerson 1

             Abstract                    will separate the data. Which of these will be found by
   Can a neural network minimizing cross-entropy           gradient descent?
   learn linearly separable data? Despite progress
   in the theory of deep learning, this question re-
                                 Our work addresses these questions as follows.
    ...