L7 - Convolutional Neural Networks

Question 1

What is a convolution layer in NNs?

Answer 1

A layer in the nework made up of n fixed size kernels.

Question 2

What is a kernel?

Answer 2

A matrix whos application as a mask will modify an image.

sometimes called a filter

Question 3

How do you use a kernel?

Answer 3

“Scan” the kernel accross the input matrix.
At each location take the sum of all the kernel values multiplied by the input values in that scan.
The output will be a matrix of these sums.

Question 4

What is pooling used for in NNs?

Answer 4

Pooling is used to down sample feature maps.
Eg maxpooling where you take only the maximum value of each patch.