The fields of machine learning and deep learning sometimes seem indistinguishable from each other. But they differ in that the second shows greater sophistication.
Artificial intelligence has become the new fetish term of the technology sector. At least it seems that this will be so during 2017, although the magnetism of the word may go further. Apart from the lexicon, the truth is that technology offers great opportunities. But it does so in more concrete ways than the breadth represented by the concept of artificial intelligence.
Two of the terms that lately sound associated with artificial intelligence are machine learning and deep learning. Both have their translation in Spanish (machine learning and deep learning), although it is not always used. They are two forms of artificial intelligence – a technology that has been working for decades-and that encompasses each other.
It can be said that machine learning has a side called deep learning. The two technologies refer to self-learning systems. The difference between the two lies in the learning method. Deep learning is more complex and also more sophisticated. It is also more autonomous, which means that once the system is programmed, human intervention is minimal.
Machine learning
The name of the technology is actually quite accurate. Learning automatically is the main feature that differentiates these systems from less advanced ones. Here an algorithm is provided with a set of rules to apply when it comes across the relevant data. But the software has the ability to adapt these rules and create new ones to improve your hit rate.
A system that seeks to identify dogs, for example, can ask to examine certain patterns : if it is a quadruped, if it has hair, if it has nose, two eyes and ears. A whole series of requirements that dogs normally meet. From there the algorithm is trained. A large number of images of dogs and other animals are introduced, indicating in each case whether or not it is a canid. This allows the software to invent new rules for itself, in order to facilitate the discovery of dogs in unknown photographs.
Deep learning
The difference between machine learning and deep learning is that the second technique brings learning to a more detailed level. The system in this case v a by layers or neural units. In fact, the functioning of these algorithms tries to mimic that of the brain.
In deep learning each layer processes the information and gives a result that is revealed in the form of weighting. That is to say, when a unit of a system analyzes a photo in search of dogs it will conclude that the image is in a 78% probability a dog and in a 22% it is not.
The second layer that analyzes the image will combine the result obtained by the first layer with its own judgment. In such a way that the weighting will be changed and it will move to the third layer , which will also use this data to draw its own conclusion. So on and on many occasions (there are algorithms that use more than 100 layers).
What this system does is reduce the margin of error and therefore, increase the accuracy of conclusions. How do you train? Introducing an exorbitant amount of information, so that their weights are polished themselves. Thus, it will not matter if the software is presented with a photo with fog and a lame dog. He’s already clear on what a dog is and what it isn’t.
Images: Rick Payette and National Institute of Health (NIH)
