Deep learning: Myths and Truths. Learn all about it
Despite decades of development, deep learning has been boosted by the increasing processing power of computers and the immense amount of data available, which has allowed it to obtain some achievements that were unthinkable years ago.
Deep learning is at the heart of what we know today as artificial intelligence. However, like everything that is talked about a lot, deep learning has formed a series of myths that do not always obey reality.
In this article we are going to see, very briefly, what some of these myths are, and what some of their truths are. Let’s go!
6 myths and truths of deep learning
– Deep learning is very recent
This is a myth, although it is understandable that it has spread. The first applications created on the basis of deep learning date back to the 1950s, as early as, for example, 1952, when Arthur Samuel wrote a program capable of learning to play checkers. For decades, however, this field of computer science was restricted to the more specialized circles.
The main reason, probably, was its difficulty in obtaining results in practice. In short, we can say that deep learning requires a considerable computing capacity to achieve useful results in a reasonable period of time, a power that has only been available in recent years. In addition, the availability of processable data in the second half of the twentieth century was much lower than today’s, which also explains its current explosion.
– The data is the “fuel” of deep learning
It’s a reality. As we have already said, and together with the greater computing power, one of the reasons why automatic learning has gained relevance in recent years is the greater availability of a massive amount of data, what we know as “Big Data”.
However, deep learning systems not only need a large amount of data, but these have to do with the results you want to obtain. Companies developing deep learning solutions therefore place particular emphasis on the importance of the quality of the pre-selection of the data with which the machine learning systems are “fed”. We will talk more about it in the next point.
– Deep learning only needs a large amount of data
As we mentioned earlier, this case is about a myth. Although Big Data is essential to deep learning systems, the quality of the data supplied to it will be even more important.
Thus, an automatic learning system can offer unsatisfactory results if the previous data “pre-selection” is of poor quality. If, for example, a deep learning system is intended to “predict” the results of an election, it will be useless if we feed it with the results of the football league and not with other more relevant data, such as electoral surveys or results of previous elections.
– Deep learning can be more “intelligent” than human beings
It is not unthinkable that this may be the case in the future, but for the time being it is a myth.
Although the power of deep learning to find correlations can exceed, even by far, that of the human being, that does not mean that it can draw intelligent conclusions. Automatic learning allows patterns to be detected, but these will not always imply causality or necessity, but may be due to spurious or irrelevant reasons. For this reason, after the detection of correlations, the intervention of human beings is usually still necessary, who assess the results obtained.
This is especially necessary in some sensitive activities in which deep learning is already at work, such as medical diagnosis. In this area machine learning can give hopeful results, but it is still very necessary the supervision of the human professional, who determines the value of the results obtained and discards inconsistencies.
– Deep learning is like human learning
Although they have certain similarities, for the moment it is a myth.
On the one hand, this statement cannot be made because we do not yet know how the human brain works in sufficient detail to state that an automatic learning system is the same as that of a human being.
On the other hand, what we do know is that the human brain is much more efficient in its learning procedures than the best of machine learning systems. The brain of a young child is capable of identifying an animal, for example, after having seen it a few times, in person or in photography, while a machine learning system would have to “view” thousands of photographs of the same type of animal in order to identify it as such in the following images shown to it.
– Deep learning is one of the technologies of the future
Even if we don’t have a crystal ball to predict the coming years, it’s probably a reality.
The potential of deep learning has been demonstrated in recent times, with increasingly promising results. A large number of companies are already using deep learning solutions, and the boom is expected to be even greater in the coming years. Thus, although it will not be easy for deep learning to comply with the high “hype” that has been built around it, it is quite likely to occupy a place among the relevant technologies in the near future.
So far we have seen 6 myths and truths of this technology. And now, what if you get to know a monitoring software called Pandora FMS?
El equipo de redacción de Pandora FMS está formado por un conjunto de escritores y profesionales de las TI con una cosa en común: su pasión por la monitorización de sistemas informáticos.
Pandora FMS’s editorial team is made up of a group of writers and IT professionals with one thing in common: their passion for computer system monitoring.