1 of 48

Before we begin…

bit.ly/ws-pt-android

apple.co/3SzO2vD

2 of 48

By: Miguel Angel Bravo (Mackaber)

Build your own AI/ML app from zero

@Midjourney prompt:  "A prototype app for artificial intelligence"

3 of 48

Agenda

  • Who Am I?
  • Why?
  • Intro Functional Prototypes/MVP’s
  • Intro AI/ML/DL
  • Google’s Teachable Machine
  • PlayTorch
  • Extras

4 of 48

Who Am I?

Miguel Angel Bravo (Mackaber)

  • Full-Stack Developer @ Encora MX (Ruby on Rails)
  • Programming Professor @ Dev.F
  • Master’s in Computer Science
  • Hackathon Aficionado
  • InfoSec Hobbyist
  • Likes: Cooking, Video games, Movies
  • Ruby/Java/JS/Python/React/SQL
  • AI (ML & DL)/ Bioinspired Algorithms
  • I break Stuff

5 of 48

Why?

​

​

6 of 48

By: Miguel Angel Bravo (Mackaber)

Build your own AI/ML app from zero

@Midjourney prompt:  "A prototype app for artificial intelligence"

7 of 48

By: Miguel Angel Bravo (Mackaber)

Build your own AI/ML app from zero

@Midjourney prompt:  "A prototype app for artificial intelligence"

8 of 48

9 of 48

ML Apps, just like iPhone early apps!

10 of 48

Intro: Functional Prototypes/MVP’s

11 of 48

The app… MVP/PoC/Toy/Prototype

  • PoC: Proof of Concept
  • Toy Example: An example that can be used to “play”, (has exposed variables which can be easily changed)
  • Functional Prototype: The main Concept works, everything else can be faked (hardcoded)
  • MVP: Minimum Viable Product; The minimal application in order to satisfy the user’s need

12 of 48

Let’s make noodle soup!

13 of 48

Cooking levels

Level 0

Level 1

Level 2

Level 3

14 of 48

Prototyping levels

Level 0

Level 1

Level 2

Level 3

15 of 48

Prototyping levels

Level 0

Level 1

Level 2

Level 3

No-Code

Low-Code

Frameworks

Languages

16 of 48

AI? ML?DL?...

17 of 48

AI Basics

  • AI: Artificial Intelligence
  • AGI: Artificial General Intelligence
  • ML: Machine Learning (statistics)
  • DL: Deep Learning
  • NN: Neural Network
  • DNN: Deep Neural Network
  • CNN: Convolutional Neural Network
  • RNN: Recurrent Neural Network

18 of 48

AI?, Machine Learning, Deep Learning?

http://www.gogeometry.com/software/ai/artificial-intelligence-mind-map-approaches.html​

19 of 48

20 of 48

21 of 48

Machine Learning basics

22 of 48

https://www.youtube.com/watch?v=cNxadbrN_aI

23 of 48

Machine Learning basics

24 of 48

Bias vs Variance

25 of 48

DNN Architecture

https://medium.com/analytics-vidhya/understanding-basics-of-deep-learning-by-solving-xor-problem-cb3ff6a18a06

26 of 48

Build your DNN from zero!

27 of 48

Build your DNN from zero!

…In Excel (or any spreadsheet)

https://docs.google.com/spreadsheets/d/1ivvyvezspz01dRGAIBzf-ZEq3LsRIiX4h_u2C_UBprM/edit?usp=sharing

28 of 48

29 of 48

30 of 48

Hyperparameters

  • Epochs: Number of times (iterations) to take the “lesson”
  • Batch size: number of samples per iteration
  • Learning rate:
    • Too few: Not learning
    • Too much: Memorizing (Overfitting)

​

31 of 48

Image Classification using

​

Google Colab: bit.ly/tl-colab

(Don’t forget to turn on the GPU!)

MobileNet

32 of 48

Image Classification using Pytorch

  1. Gather data train/validation
  2. Setup the data
  3. Train (Finetune/Transfer Learning)
  4. Transform into Playtorch format

33 of 48

  1. Get Data

34 of 48

35 of 48

or…

36 of 48

Keywords: image, scrap

37 of 48

https://github.com/MehediH/Bulksplash

38 of 48

2.Setup the data

39 of 48

3. Train!!!

40 of 48

4. Transform into Playtorch format

https://playtorch.dev/docs/tutorials/prepare-custom-model/

41 of 48

Snack link: bit.ly/snack-IC

42 of 48

Image Segmentation using

​

Google Colab:

(Don’t forget to turn on the GPU!)

DeTR

43 of 48

Image Segmentation

  1. Label data (Hasty.ai) Keywords: ”Coco Labeler”
  2. Setup Data
  3. Train (FineTune DeTR)
  4. Transform into Playtorch format

44 of 48

Snack link:

Based on:

45 of 48

Extras

46 of 48

Extras

(CARYKH)

47 of 48

Questions?

​

48 of 48

You are awesome!

❤

Thank you!

Contact:

​

Just Google: “Mackaber”

https://bit.ly/encora-ai-ppt