Skew Orthogonal Convolutions

                        Sahil Singla 1 Soheil Feizi 1

            Abstract                adversarial robustness (Cisse et al., 2017; Szegedy et al.,
                                2014) and interpretable gradients (Tsipras et al., 2018). The
  Training convolutional neural networks with a       Lipschitz constant also upper bounds the change in gra-
  Lipschitz constraint under the l2 norm is useful for    dient norm during backpropagation and can thus prevent
  ...