1. Home
  2. Training Library
  3. Big Data
  4. Courses
  5. Introduction to Data and Machine Learning

Real World Applications

Developed with
Catalit

The course is part of this learning path

play-arrow
Start course
Overview
DifficultyBeginner
Duration37m
Students488
Ratings
4.9/5
star star star star star-half

Description

In this Zero to Deep Learning course has been expertly created to provide you with a strong foundation in machine learning and deep learning. So whether you're just starting out with your practice of machine learning or you're a more experienced data scientist that is adding deep learning to the mix, this course will provide the necessary skills to serve as a solid foundation for you to continue learning after the course has been completed.

The course will start out covering simple data and structured data, then moving onto images, with sound, with text, and more complex data where deep learning comes to life. By the end of the course, you'll be able to recognize which problems can be solved with deep learning and organize data in a way that can be used by a neural network. Understanding and learning how to build a neural network model, including fully connected, convolutional, and recurrent neural network and train a model using cloud computing, all within this course.

This course is made up of 5 cohesive lectures that start off the journey into Data and Machine Learning with Cloud Academy.

 

Learning Objectives

  • Learn the key principles of data and machine learning
  • Come away with a strong foundation of the subject in order to develop new skills further
  • Understanding and learning how to build neural network models

Intended Audience

  • No prior knowledge of data and machine learning required

 

 

Transcript

Okay guys, let's get started with some applications of deep learning. Deep learning is everywhere. Literally every day there is a new application coming out. You read an article, scientific papers, news, vlogs, everything. Everybody is talking about deep learning. So in this very short video, I want to give you a glimpse on what's possible with neural nets. By no means is this going to be exhausted, but one thing I want to convince you is deep learning is more than just supervised machine learning. Although we'll introduce it as an extension of machine learning, you'll see that there is much more that you can do with neural nets than just classifying data. I'll cover a few applications and I'll make sure I highlight how this application refers to a specific data type and also to the type of industry it's from. And as you can see, there are applications of deep learning across many, many industries. So let's start with one of the most famous ones, the one that you may be very familiar with: object recognition. Neural nets are by now really good, much better than humans, at recognizing objects in images given that they've been previously trained on a large data set of images. An extension of that, which I find pretty cool, is the generation of captions from images. So the input here is an image but the output is actually a text string describing what's happening and, like, as you can see in this example, "man in black shirt is playing guitar". What are some applications of these discoveries and these inventions? Well for example, a few years ago, Facebook published their Deep Face paper where they applied image recognition to recognizing faces which is a pretty interesting application in many ways. A probably more useful application is captioning for the blind. Facebook applied the same technology for blind people to be able to know what's in images or what's in front of them using a camera which I find really, really awesome.

 Another company called Yelp, they use deep learning to classify images of food and venues to offer their users a better experience. Let's move on to a different type of data. What about sound? Many companies have come out with deep learning-based algorithms that essentially take sound as input and are able to recognize words and therefore convert speech to text. So by the research was one of the pioneers in this, Google followed immediately and by now pretty much any large company has a speech recognition system that is deep learning-based. These companies also offer APIs for you to try so make sure to check the references at the end of this video if you want to go ahead and try one of their speech recognition engines. Another very cool application is language translation. Google Translate is the first who offered that but you can find it in other websites. Essentially the idea is you have pairs of languages, for example, here you have English and German or English and French and the neural net learns to translate in a way that it's much more natural and better than previous technologies. A mashup you can do with these technologies is you can have real-time translation in an app. For example, from an image, as you can see here, or even for sound where you combine... this is an interesting combination of text-to-speech and automated translation so you can have Skype calls or you speak in one language and the person on the other side hears what you said in their own language. Interesting combination of deep learning on text and on sound. 

Another interesting application is the Smart Reply feature from Google Inbox. This uses two neural nets: one small neural network to decide whether or not they can provide a Smart Reply and, if so, it will engage a deeper, bigger network to actually generate the Smart Reply. Also with text, this is an interesting article that shows how you can generate clickbait titles with RNNs. Getting onto more scientific breakthroughs, here's an application of deep neural nets to automate diagnoses of retinal disease. Retinal disease is a terrible illness and the ability to automatically assess retinal disease is a breakthrough. The interesting aspect of this article is that it uses a pretrained convolutional neural net called Inception and we'll use it in the course , too. So if you want to build your own retinal disease detector, you'll be able to do it at the end of the course. Here's another medical application diagnosing skin cancer using deep neural nets and that network is trained to recognize if a mole on your skin is dangerous or not. Going on to industry application, here is an application also from Google where they applied deep learning to controlling the cooling system of their data center and they obtained a 40% reduction in cost using deep learning. Now if you can think of the scale of this company, this means billions of dollars saved every year thanks to an algorithm. In agriculture, there is a bunch of companies applying deep learning to several aspects from monitoring crops, to controlling equipment on the field, to predicting and optimizing yields. One interesting application because you'll be able to build it if you want is this application from a guy that built a cucumber sorter for his family farm. The cucumber sorter is trained in recognizing good cucumbers from bad cucumbers and is able to sort them. 

Self-driving cars, I'm sure you've heard of them, now many companies coming out with applications of deep learning for autonomous driving and I'm sure this is gonna be one of the big applications in the next few years. Go used to be considered an intuitive game where only humans could triumph. It's not so anymore. Last year, DeepMind's AlphaGo won the human champion of this game. Also, a game of poker recently was field for deep neural net to play with and that's interesting because playing poker involves bluffing and so the neural net had to learn to bluff in order to win champions. Video games also are a domain for computers to play. There's a domain of deep learning called reinforcement learning where the network learns to play in an environment where there is a reward like a video game. Also, the arts. People are experimenting with what's possible with deep learning. Here's an example of machine generating music after listening to Bach. And here's an example I'm sure you've heard of, style transfer where the style of a painting is transferred to another painting. So these are just a few applications of deep learning. I hope I've managed to convey a few important messages. First of all, deep learning is much richer and bigger than machine learning itself. Although it's part of machine learning, it's still a technique that learns from data. Deep learning is here to stay. It's such a revolutionary and profound breakthrough that I'm sure will see many, many application in the coming years and I hope you're excited that you will learn to build some of these during the course. Thank you for watching and see you in the next chapter.

About the Author

Students1345
Courses8
Learning paths3

I am a Data Science consultant and trainer. With Catalit I help companies acquire skills and knowledge in data science and harness machine learning and deep learning to reach their goals. With Data Weekends I train people in machine learning, deep learning and big data analytics. I served as lead instructor in Data Science at General Assembly and The Data Incubator and I was Chief Data Officer and co-­founder at Spire, a Y-Combinator-­backed startup that invented the first consumer wearable device capable of continuously tracking respiration and activity. I earned a joint PhD in biophysics at University of Padua and Université de Paris VI and graduated from Singularity University summer program of 2011.