Unit 2 Convolutional Neural Network

Question 1

What is convolutional neural network?

Answer 1

Type of artificial neural network which is also called as ConvNets, mainly used for image classification and image processing it consists of a convolutional layer that has different operations than normal hidden layers in ANN

Question 2

Components of CNN:

Answer 2

1. Convolutional layer
2. Pooling layer
3. Fully connected layer

Question 3

Components required by a convolutional layer:

Answer 3

It requires input data, a filter, a feature map

Question 4

What is a filter?

Answer 4

It is a matrix containing random numbers and of fixed size usually 3x3
Also the smallest unit of a convolutional layer which is responsible for the processing of images to detect edges and different patterns like circles, square etc

Question 5

Working of convolution layer (Convolution Operation):

Answer 5

Convolution layer receives the input images and the filters present in the convolution layer process this images to obtain different patterns like edges circles vertical sections etc.

Question 6

What is meant by ‘convolved’?

Answer 6

The filter is slider over the entire image and a feature map is generated which shows all the patterns that are observed in the input images

Question 7

What is parameter sharing?

Answer 7

Parameter sharing means using same parameters for more than one functions or neurons. This is the memory consumption and also there are fewer updations of the weights and biases

Question 8

What is equivariant to translation?

Answer 8

Property that convolution layers have due to parameter sharing
Equivariant means that if the input changes the output changes the same way.

Question 9

What are pooling layers?

Answer 9

The conduct dimensionality reduction that is reducing the number of parameters in the input. Similar to convolution layer, filter sweeps across the entire image but the differences that the filter does not have any weights

Question 10

Types of pooling layer:

Answer 10

1. Max pooling
2. Average pooling

Question 11

What is Alex-net?

Answer 11

• First large-scale CNN
• Used on large image datasets
• Sate-of-the-art results at the time of publication
• Used GPUs for training
• Used to win in ImageNet Large Scale Visual Recognition Challenge (ILSVRC)

Question 12

Activation functions used in Alex-net

Answer 12

Activation functions used in hidden layers are ReLU and in output layer Softmax is used

Question 13

Alex-net composed of:

Answer 13

5 convolutional layers with combination of max pooling layers, 3 fully-connected layers and 2 dropout layers

Question 14

Total number of parameters used in Alex-net

Answer 14

Around 60 million