Deep Learning DIY lectures

class: center, middle

# Lecture 1:
### Introduction to Deep Learning 
### ... and your setup!

Marc Lelarge

---
# Goal of the class

## Overview

- When and where to use DL
- "How" it works
- Frontiers of DL

## Deep Learning .red[Do It Yourself!]

- Fast evolving field with a lot of ressources available online
- Use a neural network right away, get results and then understand it!
- .red[This course requires your active participation]

---
# If you are (already!) lost

- Course website:
<br>
<http://www.dataflowr.com>

- Course GitHub:
<br>
<https://github.com/mlelarge/dataflowr>

---
# What is Deep Learning

### Typical ML system

.center[
          <img src="images/image_ml.png" style="width: 670px;" />
]

.credit[Slide credit: O. Grisel and C. Ollion]

---

# What is Deep Learning

### Typical ML system

.center[
          <img src="images/image_ml_2.png" style="width: 670px;" />
]

.credit[Slide credit: O. Grisel and C. Ollion]

---
# What is Deep Learning

### Deep Learning system

.center[
          <img src="images/image_dl.png" style="width: 700px;" />
]

.credit[Slide credit: O. Grisel and C. Ollion]

---
# Why Deep Learning Now?

- Five decades of research in machine learning

- .grey[CPUs/GPUs/storage developed for other purposes]

- .grey[lots of data from “the internet”]

- .grey[tools and culture of collaborative and reproducible science]

- .grey[resources and efforts from large corporations]

---
# Why Deep Learning Now?

- Five decades of research in machine learning

- CPUs/GPUs/storage developed for other purposes

- .grey[lots of data from “the internet”]

- .grey[tools and culture of collaborative and reproducible science]

- .grey[resources and efforts from large corporations]

.center[
<img src="images/gpu_tpu.png" style="width: 450px;" /><br/><br/>
<small>_GPU and TPU_</small>
]

---
# Why Deep Learning Now?

- Five decades of research in machine learning

- CPUs/GPUs/storage developed for other purposes

- lots of data from “the internet”

- .grey[tools and culture of collaborative and reproducible science]

- .grey[resources and efforts from large corporations]

.center[
<img src="images/imagenet.png" style="width: 450px;" />
]

---
# Why Deep Learning Now?

- Five decades of research in machine learning

- CPUs/GPUs/storage developed for other purposes

- lots of data from “the internet”

- tools and culture of collaborative and reproducible science

- resources and efforts from large corporations

.center[
<img src="images/frameworks.png" style="width: 500px;" /><br/><br/>
]

---
# DL Today: Vision
###Object detection and segmentation

.center[
<img src="images/pinheiro.png" style="width: 500px;" />
]

.credit[Pinheiro et al., arXiv:1603.08695]

---
# DL Today: Vision for control

<div style="text-align:center;margin-bottom:30px">
  <iframe width="560" height="315" src="https://www.youtube.com/embed/umRdt3zGgpU?rel=0&showinfo=0" frameborder="0" allow="autoplay; encrypted-media" style="max-width:100%" allowfullscreen="">
  </iframe>
<br/>Quadcopter Navigation in the Forest
<br/>Giusti et al, http://rpg.ifi.uzh.ch/docs/RAL16_Giusti.pdf
</div>

---
# DL Today: NLP, translation

.center[
<img src="images/nlp.png" style="width: 700px;" />
<br/>Wu et al., arXiv:1609.08144
]

---
# DL Today: NLP, question answering

.center[
<img src="images/nlp2.png" style="width: 720px;" />
]

.credit[Slide credit: O. Grisel and C. Ollion]

---
# DL Today: Auto-captioning

.center[
<img src="images/vinyals.png" style="width: 760px;" />
<br/>Vinyals et al., arXiv:1411.4555
]

---
#DL Today: Speech-to-Text

.center[
<img src="images/speech.png" style="width: 780px;" />
]

.credit[Slide credit: O. Grisel and C. Ollion]

---
# DL Today: Style transfer

.center[
<img src="images/transfer.png" style="width: 500px;" />
<br/>Gatys et al., arXiv:1508.06576
]

.credit[from github.com/fzliu/style-transfer]

---
# DL Today: Generative models

.center[
<img src="images/nvidia_celeb.jpg" style="width: 350px;" />
<br/>Sampled celebrities [Nvidia 2017] 
]

--
<br/>

.center[
<img src="images/stackgan.jpg" style="width: 600px;" />
<br/>StackGAN v2 [Zhang 2017] 
]

.credit[Slide credit: O. Grisel and C. Ollion]

---
# DL Today: Generative models

<div style="text-align:center;margin-bottom:30px">
  <iframe width="560" height="315" src="https://www.youtube.com/embed/G5JT16flZwM?rel=0&showinfo=0" frameborder="0" allow="autoplay; encrypted-media" style="max-width:100%" allowfullscreen="">
  </iframe>
<br/>MusicVAE, Roberts et al. arXiv:1803.05428
</div>

---
# DL for RL in games

.center[
<img src="images/games.png" style="width: 650px;" />
]

<small> AlphaGo/Zero: Monte Carlo Tree Search, Deep Reinforcement Learning, self-play </small>

.credit[Slide credit: O. Grisel and C. Ollion]

---
# Do It Yourself starts now!

### Python and Jupyter notebook

.center[
<img src="images/anaconda.png" style="width: 250px;" />
<br/>https://www.anaconda.com/download/
(Python 3.6 version)
]

### Pytorch

.center[
<img src="images/pytorch.png" style="width: 450px;" />
<br/>https://pytorch.org/
]

### Access to a GPU

.center[
<img src="images/colab.png" style="width: 250px;" />
<br/>https://colab.research.google.com/
]

---

# And now a first example!