Dropout regularization - prediction step

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
1

Hi,

Is it correct to understand that, we should turn off dropout even when predicting the accuracy of the model on the training set (i.e., when computing the training set accuracy)? I would think so but would like to confirm my understanding.

Thanks,
TK

2

How can you turn it off?

3

Hi @A_Kong,

Yes. We don’t use dropout for any prediction.

Raymond

1 Like