Chapter 5: Deep Learning for Computer Vision

Question 1

What kind of Deep learning models are most often used for machine vision problems?

Answer 1

Convolutional Neural nets or covnets

Question 2

Fundamental difference between convolutional layers and dense layers?

Answer 2

convolutional layers learn only local patterns while dense layers learn global patterns

Question 3

Key properties of covnets

Answer 3

1. the patterns they learn are translation invariant (can recognize the same pattern in different areas)
2. they can learn spatial hierarchies of patterns (start small learning simple patterns learn progressively large/more complex patterns at later layers)

Question 4

What do covnets operate over?

Answer 4

3D tensors called feature maps

Question 5

What do covnet layers produce in feature map depth

Answer 5

filters, encoding specific aspects of the input data

Question 6

Response map

Answer 6

a 2D map of the presence of a pattern at different locations in an input