On Learnability via Gradient Method
     for Two-Layer ReLU Neural Networks in Teacher-Student Setting

                      Shunta Akiyama 1 Taiji Suzuki 1 2

            Abstract                 for the generalization aspect. In this study, we tackle these
                                 two problems in a teacher-student problem with the ReLU
   Deep learning empirically achieves high perfor-       activation under an over-parameterized setting. In this set-
   mance in many applications, bu ...