There are so many important use cases for Deep Learning, that it’s impossible to produce

an exhaustive list.

Deep Learning is just getting started, and new applications pop up all the time.

Let’s take a look at some of the biggest ones today.

At this point, it should be no surprise that machine vision is one of the biggest applications

of deep learning.

Image search systems use deep learning for image classification and automatic tagging,

which allows the images to be accessible through a standard search query.

Companies like Facebook use deep nets to scan pictures for faces at many different angles,

and then label the face with the proper name.

Deep nets are also used to recognize objects within images which allow for images to be

searchable based on the objects within them.

Let’s look at an example application – Clarifai.

Let’s load Clarifai in a browser.

Here is the URL, which you'll also find in the video description below.

Clarifai is an app that uses a convolutional net to recognize things and concepts in a

digital image.

Let’s take a look.

Right in the middle of the page you have the demo button.

Lets click that.

It takes you to part of the webpage where you have the demo.

You have two choices - a) either choose a URL where the image is located, or b) load

the digital image yourselves if you have it on file.

I'm going with choice b) - loading an image; I am in the right folder now and am going

to select the first one.

When I select an image, it wants me to go through a verification process.

In this case, it wants me to select all the squares that have a gift box, so I'm gonna

go through and do that.

This changes every time btw - you can have different tests.

Its come back and you see the predictions.

Firstly, it says there's no person, it expected to find a person but there weren’t any so

it identified that as a pattern for this one, which is cool!

The other predictions are "tableware", "indoors", "party", "fashion" etc.

So this is the list of tags its associated with this image.

If you scroll down, it shows a list of example images and the items in them.

Like the first one with a coffee and croissant, which I think is cool.

If you go to the one with the concert, its tagged it pretty accurately with "concert",

"band", "singer" etc.

You also get similar images.

I'm going to pick another one, this time of a county fair.

Again it goes through the same verification process - this time it wants me to pick images

with cars.

Ok - it came back and gave me some tags.

It recognized a Ferris wheel, and though carousel is only partly visible to the left, it still

picked it out!

It also picked out the word "fun".

Also, the images it suggested as similar are accurate - they are virtually identical to

the one I picked.

Further, it presents the same example images as the last time.

So there you have it, a demo of object recognition using Clarifai.

Other uses of deep learning include image and video parsing.

Video recognition systems are important tools for driverless cars, remote robots, and theft

detection.

And while not exactly a part of machine vision, the speech recognition field got a powerful

boost from the introduction of deep nets.

Deep Net parsers can be used to extract relations and facts from text, as well as automatically

translate text to other languages.

These nets are extremely useful in sentiment analysis applications, and can be used as

part of movie ratings and new product intros.

Here is a quick demo of Metamind - an RNTN that performs sentiment analysis.

Let’s load Metamind in a browser.

Here is the URL, which you'll also find in the video description below.

Metamind is an app by Richard Socher that uses an RNTN for twitter sentiment, amongst

other things.

You can search by user name, or keyword or hashtag.

I'm going to search by hash tag.

My first one's #coffee.

When you click "Classify", it first downloads the tweets which takes a little time.

It then comes back and displays you two things.

On the left, it shows you a pie chart of the 3 different sentiments - positive, negative

and neutral.

For most searches, you'll get lots of neutral comments which is natural, but here you have

more positive comments than negative - 206 vs 41, which I think is good :-)

On the right, it also lists some example comments classified as positive, neutral and negative.

Let’s search a different one - #holidays.

Not surprisingly, you find a ton more positive comments about the holidays.

In this case, if you look at the example, even the negative ones are light-hearted.

So there you have it, a demo of twitter sentiment analysis using Metamind.

Even recurrent nets have found uses in character-level text processing and document classification.

Deep nets are now beginning to thrive in the medical field.

A Stanford team used deep learning to identify over 6000 factors that help predict the chances

of a cancer patient surviving.

Researchers from IDSIA in Switzerland created a deep net model to identify invasive breast

cancer cells.

Beyond this, deep nets are even used for drug discovery.

In 2012, Merck hosted the Molecular Activity challenge on Kaggle in order to predict the

biological activities of different drug molecules based solely on chemical structure.

As a brief mention, this challenge was won by George Dahl of the University of Toronto,

who led a team by the name of ‘gggg’.

But one crucial application of deep nets is radiology.

Convolutional nets can help detect anomalies like tumors and cancers through the use of

data from MRI, fMRI, EKG, and CT scans.

In the field of finance, deep nets can help make buy and sell predictions based on market

data streams, portfolio allocations, and risk profiles.

Depending on how they’re trained, they’re useful for both short term trading and long

term investing.

In digital advertising, deep nets are used to segment users by purchase history in order

to offer relevant and personalized ads in real time.

Based on historical ad price data and other factors, deep nets can learn to optimally

bid for ad space on a given web page.

In fraud detection, deep nets use multiple data sources to flag a transaction as fraudulent

in real time.

They can also determine which products and markets are typically the most susceptible

to fraud.

In marketing and sales, deep nets are used to gather and analyze customer information,

in order to determine the best upselling strategies.

In agriculture, deep nets use satellite feeds and sensor data to identify problematic environmental

conditions.

Which of these deep learning applications appeals to you the most?

Please comment and share your thoughts.

In the next video, we’ll take a look at the main ideas behind a Deep Learning Platform.