SPEAKER 1: Hey, everybody.

Welcome to the ML & AI sandbox here at I/O 2019.

So in the keynote, we saw some amazing stuff

that you can accomplish with on-device ML.

And I'm here with TensorFlow Lite,

showing developers how they can build that same stuff

themselves.

So you've got everything from object detection

to image classification and voice recognition.

And we're showing people how they can deploy that

down to devices that are really tiny

and run for weeks on a single battery.

So to really show you what's possible with TensorFlow Lite

on-device, we built this amazing experience called Dance Like.

SPEAKER 2: OK.

Behind me is Dance Like.

It's super fun application using TensorFlow Lite, which

is TensorFlow's mobile and embedded systems framework

for running machine learning.

It basically teaches people how to dance.

And so it does this by running pose segmentation on the GPU.

We have a bunch of GPU-related ops

that we just released for TensorFlow Lite,

so you should check those out.

It enables you to run super fast models.

We're running two in real time and giving the user

a heap of feedback, which helps them dance better.

So to go and build really interesting applications

on mobile phones, embedded systems,

and executing this all on-device,

you can go to TensorFlow.org/Lite.

There's heaps of sample code.

There's lots of documentation and tutorials.

And you can go and build an amazing application like Dance

Like and release it to the world.

SPEAKER 3: What we're looking at here is a model called PoseNet.

PoseNet is a model that's trained

on images of human beings, and is trained

to predict their skeletal pose.

This model here is actually running entirely

in the browser with a library called TensorFlow.js.

TensorFlow.js is a machine learning library for JavaScript

that can run entirely in the browser

and use your GPU through WebGL.

Now, TensorFlow.js also runs in Node.js using the entire

TensorFlow C++ binary and all of its hardware acceleration

storage behind it.

Piano Genie is a model that runs in TensorFlow.js, entirely

in the browser as well, that generates MIDI notes

as part of a piano performance.

Now, this model is interesting because it

brings a human in the loop and lets

the human condition where the model is actually going to go.

TensorFlow.js is open source and is available online

on TensorFlow.org/js, as are multiple more examples

and demos of how to use it.

Head over there today to start building machine learning

for the browser.