The media update team take a closer look at deep learning to find out why it is so important to the future of AI.

Before you can fully understand deep learning, you have to begin with the process that came before it: machine learning. Machine learning refers to the training of machines by humans, with humans using large amounts of data to train the machine to do a specific task or return a certain result. The machine uses the data given to it as its library, applying the lessons it learns to the job at hand.

Deep learning builds on the machine’s own work

Machine learning is still, in most respects, affected by the guiding hand of humans, and deep learning takes its processes one step further. Instead of relying on task-specific algorithms, deep learning forms artificial neural networks – similar to a human brain structure – to build knowledge one layer at a time.  

Through large, connected neural networks, the performance of deep learning algorithms can increase in speed. This is because the machine is learning from its own results on a continuous basis, and, therefore, the complexity the machine is able to deal with increases. 

Deep learning in the real world

Deep learning can best be seen in Apple’s voice-guided operating system, Siri.  Through speech recognition, Siri is able to recognise its owner's voice through the synthesis of speech – the artificial copying of human speech. Siri was moved to a neural network system in 2014, and since then has seen some impressive upgrades. It has learned to identify people who are not on your contact list, but have emailed you, and other tasks learned from past experiences by joining the dots.

Photo recognition is another area where deep learning has made an impact. As mapped out by Roger Parloff on the Fortune blog site, during the training phase, a neural network is given thousands of labelled images of different animals and learns to classify them.

An unlabelled image of a dog is then shown to the pre-trained network, and starting at its first layer with different shapes and edges, by the time it reaches its top layer, it would be able to identify, to a degree of certainty, that the unlabelled image given to it is a dog.

Facebook has been utilising deep learning techniques in its facial recognition, as the platform is able to identify when someone is in a photograph, even though they have not been tagged. As deep learning systems become more advanced, expect them to become more prevalent in your day to day lives.

Want to stay up to date with the latest media news? Subscribe to our newsletter.

Machine and deep learning both play important roles in natural language processing, but what is it? Read more in our article, What is natural language processing?