Loading

React Native Machine Learning

Gant Laborde

This is a live streamed presentation. You will automatically follow the presenter and see the slide they're currently on.

Taming the Machine

Epic Machine Learning

Gant Laborde .com

AKA

AKA

@GantLaborde

infinite.red

Today

What I SHOULD talk about?

Ignite

Quick Start React Native

 

What I should NOT talk about?

tinytuba.com

What I SHOULD talk about

What I should NOT talk about

What I SHOULD talk about

academy.infinite.red

What I should NOT talk about

Crazy Machine Learning & Jokes

Machine
Learning

Why ML?

?

Why ML?

The short, short version

ML

Machine Learning

Cool and Futuristic Mobile

⚠️ Talk Warning ⚠️

Let's keep this all about visuals

⚠️ Talk Warning ⚠️

Let's keep this all about supervised learning

How does ML work?

?

How ML?

The short, short version

Neuron

The Brain

Basic ML Models

Traditional Coding

What About Images?

Image Inputs

ML and Mobile

Quick and Dirty ML

Just use services

Amazon

Microsoft

Google

and many more...

Quick and Dirty ML

Just use services

Amazon

Microsoft

Google

and many more...

Quick and Dirty ML

Cost?

  • $$$ - Software as a Service
  • No Offline Access
  • Big Brother Issues

React Native + ML Models

  • Swift Landmines

  • Xcode Madness

  • Mojave and iOS Betas

Not Alone!

Let's Play!

Seefood Founder

Over 5GB of Data

Trained an 80MB Model

Let's do some

react-native-vision

!

WE GOT HIM!

About this Presentation

AKA Why my Girlfriend

thinks I'm crazy

How does one create a Nic or Not model?

Steps to get Data

  • Scrape the internet for pictures of Nic Cage

  • Scrape the internet for pictures of Celebs

  • Code to remove faces of all images (​faceOff)

  • Manually check your data for errors

Code for above available

https://github.com/gantman/nicornot

Nic or Nickel Back?

5+ Hours!?

OF THIS!?!?!?!?!

17KB File Result!?

👎

CoreML Training Playground

How-to Machine Learning

Training a Model

NVIDIA DIGITS

Oh No!

Add a great face CNN

500k images trained

WOAH!  Looking good!

Quantize the Model Weights

227MB -> 114MB

I've probably gone too far

Also, I did it wrong

More Links

Coming Soon

.com

Thanks

Infinite Red

Thanks

Conference Organizers

Thanks

Ray!

Thanks

Nic Cage

Thanks

@GantLaborde

GantLaborde.com