C5

Question 1

Why do we need many layers in the network?

Answer 1

• in theory, on hidden layer is enough to model any function, but the number of nodes and weights grows exponentially fast
• the deeper the network, the less nodes are required to model complicated functions
• consecutive layers learn features of the training patterns, from lower layers to top layers

Question 2

convolutional layer

Answer 2

a layer of neurons that perform the same operation on fragments of the input

Question 3

feature map

Answer 3

the result of applying the convolutional layer to the data

Question 4

padding

Answer 4

artificially increasing the size of the input (eg. by adding zeros or mirror reflections) to preserve the original input size in the feature map