Deep Learning - a gentle dive by Dries Cronje

Deep Learning has the ability to disrupt and to disrupt fast. There are tools and algorithms mature enough to add value in almost any industry and you do not need a Math Ph.D to learn Deep Learning.

In this talk, I will briefly paint a picture of the exciting world of Deep Learning and then explain Deep Learning concepts using Convolutional Neural Networks as a base.

Transcript

1. Deep Learning A gentle dive
2. @dries139 SLACK braai-light
3. Autopilot
4. AlphaGo beats Go world champion AlphaGo beating Sedol
5. Image annotation "black and white dog jumps over bar." “Man in black shirt is playing guitar"
6. Real Time Translation
7. Credit: https://github.com/junyanz/CycleGAN Style Transfer
8. ??? "A woman holding a teddy bear in front of a mirror."
9. What is Deep Learning?
10. “deep learning so far has been the ability to map space X to space Y using a continuous geometric transform, given large amounts of human-annotated data. Doing this well is a game-changer for essentially every industry, but it is still a very long way from human-level AI.” François Chollet
11. What is Deep Learning
12. What is Deep Learning X Y Magic Learning phase
13. AlexNet 2012 Feature engineering Deep Learning
14. ImageNet competition 2012 AlexNet 2013 ZFNet 2014 VGG Net 2014 GoogLeNet 2015 Microsoft ResNet And this is how Deep Learning became a big thing
15. Convolutional Neural Network
16. Layered approach
17. Layered approach Layer 1 Layer 2 Layer 3
18. Layered approach a typical layer: 1) Convolutions 2) Activations 3) Pooling Layer 1 Layer 2
19. 1. Convolutions Credit: medium.com/towards-data-science
20. 1. Convolutions
21. 1. Convolutions Edge Detection as an example of a convolution being applied to an image * =
22. 1. Convolutions Striding Stride = 1 Credit: adeshpande3.github.io
23. 1. Convolutions Striding Credit: adeshpande3.github.io Stride = 1
24. 1. Convolutions Padding Credit: https://medium.com/@Synced/ Stride = 1
25. 2. Activations ReLU - non-linear activation 1) Replaces sigmoid 2) Less processing power
26. 3. Pooling Max Pooling - Reducing dimensionality Credit: cs231n.github.io
27. Choosing Hyperparameters
28. (Convolution + activation + max pooling) + fully connected + softmax Bringing it all together
29. (Convolution + activation + max pooling) + fully connected + softmax Bringing it all together
30. Softmax Credit: www.pyimagesearch.com
31. Training X Y Magic [CAT] [DOG]
32. Inference Magic [CAT]
33. Let's build something
34. Python - a 1st class citizen
35. Stay in touch @dries139 https://github.com/chasingbob/pycon2017