ML_Papers Gradient Descent Maximizes the margin of homogeneous neural networks. The Implicit Bias of Gradient Descent on Separable Data