Skip to content

Exploding Gradient

Gradient values become too large during backpropagation

Common in deep rnn → weights grow exponentially

Solutions:

  • Gradient clipping
  • lstm architecture
  • Proper weight initialization

References


#ml-notes

On this page