3

Question 1

What is Convolutional Neural Network (CNN)?

Answer 1

Deep learning model specifically designed for processing grid-like data, such as images, videos, or time-series data. CNNs are highly effective in tasks involving visual data, like image classification, object detection, and segmentation, because they can automatically capture spatial hierarchies and patterns from raw pixel data

Question 2

How does Convolution work?

Answer 2

We take one layer, pass it through the filter and get the Activation map. Do it for all z dimensions (x-y-z)

Question 3

What are parameters of the Convolutional map and what they do?

Answer 3

Kernel size (k) - defines the size of the filter
Stride (s) - how many pixels to move?
Padding (p) - how much extend the borders

Question 4

How to calculate the dimensions of the feature map after going through convolution?

Answer 4

(N-F/S) + 1

Question 5

How to calculate the number of parameters in the feature map?

Answer 5

(filter’s lenght * filter’s height * image’s width +1) * number of filters

Question 6

Why use CNNs instead of Fully-Connected layers?

Answer 6

CNNs have less parameters
CNNs exploit the spacial structure of the image
Better generalization due to shared weight structure.

Question 7

How Pooling and Max Pooling works?

Answer 7

Pooling - reduces the size of the image (resolution) without adding additional learnable parameters
Max Pooling - find a maximum value of a filter. It doesn’t have any parameters