[MUSIC] So we saw that deep learning had
a tremendous part in the ImageNet competition. Which allowed them to take 1.5 minute
image string deeply on your network and get amazing performance to predict one
of a thousand different categories. So let's go ahead and
show you a little demo of what kind of categories we're talking about and
how cool the predictions were. So here's an example. It was the AlexNet frame
on that ImageNet data set, which we then employed as a service
that can be queried from this website. And so every time I click on an image
it gets sent to that service which actually runs on a GPU, so it's
fast and it comes back for prediction. So if I click on this
particular image here, it gets sent to a service that
actually hosts in on Amazon AWS. It comes back for prediction here. It's hidden, but when I click on it,
it tells me what prediction is. So if I show you this image, it might
be unclear what that image is, but if I click on it, it says parking meter,
it turns out to be the right label. The second best prediction was padlock,
Which you can see kind of a padlock. The parking meter, you got it right. So that's really quite cool. Let me show you another example. For example, this one. It get shipped off to the service on
Amazon WS, comes back for prediction. And here my prediction, screen, monitor,
or [INAUDIBLE] it says it's a monitor, but I don't know what the difference between a
screen and a monitor is, but that's okay. So there's various images here,
I'm just gonna click on a few. So for example, if I click on this
one over here, it gets sent up. It's really sure,
it really thinks it's a spoonbill, it turns out to be a spoonbill,
which is great. Lastly I'm gonna click on
this one over here and that image gets sent to that service
uses a deep learning for GPU and it says it's a beer bottle or pop bottle,
the true label is beer bottle. And now this all image is in
the original ImageNet data set, I'm sure an image that was not in
the original ImageNet data set, I click on this one here,
it gets sent to that service. On the AWS that we're hosting there. Comes back for prediction, it says,
Labrador retriever, this is my dog. This is the lab in
>> [LAUGH] And this is the dog over here, in Dato. And so, as you can see, even for images
that were not in the original data set, you can still get pretty
interesting predictions. Now in your capstone,
you want to build a service now not for predicting images here but for
recommendations with deep learning, for product images and text and
host it as a service. And you'll be able to get a website like
this, that anybody can play with, use, and really see the power of the machine
learning that you've been learning about. [MUSIC]