A Beginner's Guide To Deep Learning

Digiryte logo 180x180 round fit
Digiryte April 16, 2019 12:15 3 min read 612 views so far!

What is Deep Learning?

Where machine learning is a subset of AI, deep learning is a subset of machine learning. Similar to how we learn from experiences, deep learning repeats the same task, each time modifying itself to get a better result. Deep Learning is the backbone of AI, with various “deep” layers of neural networks enabling learning.

The layers of neural networks mimic that of a brain, except deep learning can “learn” a lot faster than a human can. Google demonstrated the potential, with the right infrastructure, with a neural network with more than a billion connections.

How does Deep Learning work?

Deep Learning uses neural networks. Those neurons are sorted into three different types of layers.

  • Input Layer

  • Hidden Layer(s)

  • Output Layer

Data enters through the input layer, which separates the information given into specific neurons, and to the next, hidden layers.

Hidden Layers do the mathematical computations. One of the obstacles that programmers face when building deep learning networks is deciding on how many layers to put in, and how many neurons will be in each layer.

The output layer collects all the computations made in the hidden layers and outputs it in the way we can understand it.

The “Deep” in Deep Learning refers to having more than one hidden layer.

Above is a very quick and basic explanation of how Deep Learning works, Click here for more in-depth visualisation.

What is Deep Learning being used for now?

As mentioned before, Deep Learning is the backbone of AI. Everything that AI is able to do, is because of Deep Learning technologies.

For the most part, deep learning is used to process large amounts of information quickly, although there is some surprising application to JPEG image compression. At the moment, research into it has shown that they are able to reduce VGG-16 from 552mb to 11.3mb with no accuracy loss.

Adding colour to black and white images and videos is one application that is using Deep Learning AI. In practice that would take someone a very long time by hand, can now be done automatically with Deep Learning. The results are very accurate and impressive.

Other applications include facial recognition, chatbots/service-bots and translations for languages.

Skype has shown how deep learning can be used to help translate over 60 languages in real time. It has shown that the programme can also translate 10 different languages through a voice call, in real time.

Where will Deep Learning take us?

Deep Learning is a rapidly growing field, with new research coming out all the time about what has been done, and what can be done soon.

It is difficult to keep track of all the news coming out about it but is important to keep an eye out for technologies that could expand and change the science of Deep Learning, Machine Learning and AI.

Some of the changes that might come soon are the compression of neural networks. Deep Compression, as it is being called, works similarly to JPEG image compression. At the moment, research into it has shown that they are able to reduce VGG-16 from 552mb to 11.3mb with no accuracy loss.

A lot of money is being put into Deep Learning. Google is said to have paid $400 million for DeepMind, a London based AI company, specialising in deep learning research. And Peter Lee, head of Microsoft Research has said that a “world class deep learning expert” is capable of demanding a seven-figure salary.

This kind of movement in the world of AI and Deep Learning, surely something big is on the horizon.

Know how Deep Learning can help your business Click Here