Convolutional neural networks

Question 1

What does CNN stand for in the context of neural networks?

Answer 1

Convolutional Neural Network

Question 2

True or False: Convolutional layers are primarily used for processing sequential data.

Answer 2

False

Question 3

What is the primary purpose of a convolutional layer in a CNN?

Answer 3

To extract features from the input data.

Question 4

Fill in the blank: A convolutional neural network typically consists of convolutional layers, pooling layers, and ________ layers.

Answer 4

fully connected

Question 5

What operation is performed by the pooling layer in a CNN?

Answer 5

Downsampling the feature maps.

Question 6

What is a common type of pooling used in CNNs?

Answer 6

Max pooling

Question 7

What does a filter (or kernel) do in a convolutional layer?

Answer 7

It slides over the input data to produce feature maps.

Question 8

True or False: CNNs are exclusively used for image data.

Answer 8

False

Question 9

What is the role of the activation function in a CNN?

Answer 9

To introduce non-linearity into the model.

Question 10

Name one commonly used activation function in CNNs.

Answer 10

ReLU (Rectified Linear Unit)

Question 11

What is the purpose of dropout in a CNN?

Answer 11

To prevent overfitting.

Question 12

How does a CNN differ from a traditional neural network?

Answer 12

CNNs use convolutional layers to process data with spatial hierarchies.

Question 13

What is the term for the process of adjusting the weights in a CNN during training?

Answer 13

Backpropagation

Question 14

Fill in the blank: The output of a CNN is often fed into a ________ layer for classification tasks.

Answer 14

softmax

Question 15

What is the significance of ‘stride’ in a convolutional operation?

Answer 15

It determines how much the filter moves across the input data.

Question 16

True or False: Increasing the number of filters in a convolutional layer increases the model’s capacity to learn features.

Answer 16

True

Question 17

What does ‘padding’ do in a convolutional layer?

Answer 17

It adds extra pixels around the input to control the spatial size of the output.

Question 18

What is transfer learning in the context of CNNs?

Answer 18

Using a pre-trained model on a new but related task.

Question 19

Which architecture is known for its use of CNNs in image classification?

Answer 19

AlexNet

Question 20

What is a common dataset used for training CNNs in image recognition tasks?

Answer 20

ImageNet

Question 21

What is the purpose of batch normalization in CNNs?

Answer 21

To stabilize and accelerate training by normalizing the inputs of each layer.

Question 22

Fill in the blank: The final layer of a CNN for binary classification typically uses a ________ activation function.

Answer 22

sigmoid

Question 23

What is the main advantage of using CNNs over traditional machine learning methods for image processing?

Answer 23

CNNs automatically learn hierarchical feature representations from raw data.

Question 24

True or False: Convolutional layers can only process 2D data.

Answer 24

False

Question 25

What is the purpose of using multiple layers in a CNN?

Answer 25

To capture increasingly complex features at different levels of abstraction.

Question 26

How can you think of a (discrete) convolution between 2 arrays?

Answer 26

Flip one of the arrays and offset them: each value of the convolution is a multiplication and sum the overlapping values.

Question 27

What are two examples of where discrete convolution can be used?

Answer 27

- Add up two discrete probability distributions - Compute the moving average of a list of numbers (average of data inside a window)

Question 28

What are some examples of discrete convolutions in image processing?

Answer 28

- Blurring - Weighted blurring (e.g. w Gaussian distributed window) - Edge detection

Question 29

What is the convolution of 2 polynomials?

Answer 29

Finiding the polynomial product of the 2 input polynomials

Question 30

What is a O(N\log(N)) way to compute discrete convolutions?

Answer 30

FFT of the two numbers, multiplication and FFT^(-1)

Question 31

How does the convolution of polynomials relate to the convolution of numbers?

Answer 31

The numbers can be treated as coefficients of polynomials, which can be sampled as many times as the degree. Multiply together the samples and solve the linear system to find the coefficients of the product polynomials. Those coefficients are the convolution of the original 2 numbers. Cost: O(N^2)

Question 32

What is the problem with processing images with fully-connected NNs?

Answer 32

That the image must be flattened, which destroys the structure and relations between pixels.

Question 33

What does it mean that NNs are hierarchical?

Answer 33

That at each layer they produce a hierarchy of features.

Question 34

How can you extract different features from an image?

Answer 34

By changing the features (kernels)

Question 35

What is learned during training of a CNN?

Answer 35

The weights of the filters.

Question 36

Why pooling?

Answer 36

Covering larger portions of the mapped images -> Getting more “macro” features.

Question 37

Why the fully-connected layer at the end of a CNN?

Answer 37

To go from a feature set to probabilities of the classes (finally apply SoftMax).

Question 38

Why does the output of a convolutional layer on a 2D image is generally 3D?

Answer 38

Because it can apply more than one filter (3rd dim = number of filters)

Question 39

What is the stride of a convolutional layer?

Answer 39

The slide step.

Question 40

What is the receptive field of a convolutional layer?

Answer 40

The locations in the previous layer that a node in the current layer is connected to.

Question 41

Why a nonlinear operation (such as ReLU) after a convolution?

Answer 41

Because it can further enhance features (e.g. by filtering out negative values).

Question 42

What is MaxPool?

Answer 42

Take only the max value of the features enclosed by the pooling window.

Question 43

When does the fully connected layer at the end NOT break the spatial structure of the image?

Answer 43

When there have been enough hierarchical transformations (convolutional layers) of the input image, which kept the spatial structure.