Tweeted By @unsorsodicorda

Third paper in a few days proving the convergence of GD and SGD for DNNs! (Well, second, actually, b/c Du et al. proof isn't valid for nonsmooth activation functions). A great month for the theory of #DeepLearning! https://t.co/LXPtWb3DWt cc @roydanroy

— andrea panizza (@unsorsodicorda) November 22, 2018