Deep learning

Deep learning is a machine learning technique whereby computers are taught to perform tasks in the same way as the human mind. It involves teaching machines knowledge through experience; in other words, learning from examples. 

For example, Deep learning is an indispensable technology for autonomous vehicles, as it enables them to recognise a road sign or distinguish between a pedestrian and a street lamp. In short, the aim is to achieve results based on knowledge that the machine did not previously possess.

Through the deep learning, a computational model performs classification tasks based on images, text or sound, achieving a level of accuracy that is virtually unprecedented and often surpasses human capabilities.

In order to train a machine using deep learning, a huge amount of labelled data is required and validated, as well as a GPU with excellent processing power to minimise the machine’s training time as much as possible.

How does deep learning work?

Computer programmes that use deep learning learn in the same way as a baby, for example, learns to recognise a dog. The algorithms in the hierarchy apply a non-linear transformation to the information they receive and use part of what they have learnt to create a statistical model as a response.

So, all you have to do is feed the programme plenty of inputs until it achieves a level of flawless accuracy. The process of providing a programme or machine with plenty of inputs is called “iteration”, and it is precisely this repetition that leads us to call this discipline “deep”.

The only limitation of the deep learning The fact is that it cannot really come up with anything on its own, but relies entirely on the data and its accuracy. Therefore, the outputs produced by a machine using deep learning will be accurate to the extent that the inputs are correct and objective. In turn, the outputs or predictions provided by a machine are limited to the inputs it has received and do not go beyond them.

What is the difference between deep learning and machine learning?

In the machine learning Traditionally, the programmer providing the inputs has to be extremely specific and practically guide the programme step by step so that it can, for example, identify an object such as a dog in an image.

For its part, in deep learning, the programme is presented with a vast amount of unlabelled and unstructured data which, through thousands of iterations, will process the data in a matter of minutes in order to identify and distinguish images featuring a dog. 

Thanks to a number of technological innovations such as the big data or the cloud computing, we can collect enough data to generate the large volume of inputs required for deep learning.

So, the deep learning and the machine learning They differ in that deep learning requires far more resources than machine learning, but it is more rigorous and faster, and does not require constant supervision by programmers.

Examples of deep learning

• Customer service: On websites, chatbots are being trained using queries and responses to improve the quality and speed of customer service.
• Text generation: Machines and programmes are learning how to mimic the grammar, vocabulary and style of a text in order to reproduce it in a completely original text.
• Military dominance: Deep learning is being used to identify safe and unsafe areas.
• Design: With deep learning, black-and-white photos can be colourised, something that used to involve a great deal of hard work.
• Medicine: In oncology, deep learning is being used to distinguish between cancerous and non-cancerous cells.