Sign in

Gradient descent provably escapes saddle points in the training of shallow ReLU networks.

Patrick CheriditoArnulf JentzenFlorian Rossmannek
Published in: CoRR (2022)
Keyphrases
  • saddle points
  • scale space
  • saddle point
  • objective function
  • training set
  • worst case
  • critical points
  • genetic algorithm
  • training samples
  • image processing
  • information extraction
  • structured prediction