Gradient flow dynamics of shallow ReLU networks for square loss and orthogonal inputs

To appear in NeurIPS 2022

Authors: E. Boursier, L. Pillaud-Vivien and N. Flammarion