#self-normalizing-networks

Question 1: Select all that apply

You replace batch normalization in a deep stack of dense layers by SELU activations and want the network to self-normalize. Which of these are required? Select all that apply.

Standardize the input features (mean 0, standard deviation 1)1
Initialize the weights with LeCun normal initialization (variance 1/fan-in)2
Initialize the weights with He initialization (variance 2/fan-in)3
Keep a batch-normalization layer after each SELU layer4

#self-normalizing-networks

Timed test