1 result

for neural-networks

• Dropout: switch neurons off in training so no circuit can lean on one
  Dropout is a regularisation technique against overfitting: during training only, randomly set the inputs and/or outputs of neurons to zero. The article's mechanism claim is that this prevents complex
  field/dropout · regularization, training, neural-networks, overfitting, dropout