chapter 7

Question 1

AlexNet

Answer 1

is a convolutional neural network for image classification. It consists of five convolutional layers followed by three fully connected layers.

Question 2

CIFAR-10 dataset

Answer 2

dataset consists of 60,000 training images and 10,000 test images, each belonging to one of the ten categories airplane, automobile, bird, cat, deer, dog, frog, horse, ship, and truck. All are color images.

Question 3

Translation Invariance

Answer 3

one of the most important characteristics of CNN, that the network will be able to recognize the target in any position even if it shifted or rotated. it is achieved by employing weight sharing between neurons as well as making them sparsely connected.

Question 4

Translation

Answer 4

is a geometric transformation, it changes the location of an object without changing its shape.

Question 5

The topology of a convolutional layer

Answer 5

• The neurons are arranged in three dimensions.
• Two of the dimensions correspond to the width and height.
• The neurons are grouped into channels or feature maps in a third dimension.
• There are no connections between the neurons within the same convolutional layer
• all the neurons within a single channel have identical weights (weight sharing), but they will receive different input values.

Question 6

Convolutional kernel

Answer 6

The operation that each neuron in a convolutional layer implements.

Question 7

Receptive field

Answer 7

The region of pixels from which a neuron receives inputs.

Question 8

Stride

Answer 8

is the number of pixels shifts over the input matrix (image).

Question 9

padding

Answer 9

The process to adding some extra pixels with zeros outside the image.

Question 10

A convolutional layer

Answer 10

consists of multiple channels or feature maps. All
neurons within the same channel share weights.

Question 11

Sparse connections

Answer 11

• reduce the total number of weights.
• reduce the number of computations.
• reduce the number of weights to store.
• reduce the number of weights to learn

Question 12

Weight sharing

Answer 12

reduces the number of unique weights and thereby reduces the number of weights to store and to learn.