1 of 17

Convolutional Neural Networks

Deep Learning

2 of 17

Main Idea

The concept is inspired by the anatomy of the visual cortex and the image representations in the brain.

3 of 17

Seminal Paper:

4 of 17

What was found:

5 of 17

How can computers represent images?

6 of 17

How can we simulate the electrical activity in the brain?

Answer: we can use numbers and special functions called “convolutions.”

Convolutions are based on image filters

7 of 17

How convolutions work:

8 of 17

Padding needs to be considered:

9 of 17

Intuitive example:

10 of 17

What happens in an activation layer:

11 of 17

Example for the neural network architecture:

12 of 17

Pooling:

13 of 17

Dropout:

14 of 17

The anatomy of a convolutional neural network:

15 of 17

Some intuition:

16 of 17

More intuition:

17 of 17

Example Applications:

  • Image analysis and pattern recognition.

  • Robots that can use image information.

  • Art: https://openai.com/dall-e-2/