#hidden-layer

Question 1: Single choice

Which statement about the hidden layers of a feedforward network is correct?

Their units have no targets in the data; they learn whatever features help reduce the output loss.1
They pass the raw input features on unchanged, without applying any learned weights or biases.2
Each hidden unit is trained against its own labelled target that is supplied with the data.3
They must contain exactly as many units as the input layer so that the matrix shapes match.4