@misc{indiciaeba376c4c135c,
  title = {Gradient descent provably escapes saddle points in the training of shallow ReLU networks},
  author = {Patrick Cheridito and Arnulf Jentzen and Florian Rossmannek},
  year = {2024},
  doi = {10.1007/s10957-024-02513-3},
  url = {https://arxiv.org/abs/2208.02083},
  note = {Source identifier: 2208.02083}
}