What is machine learning and why is it becoming so popular?1 May 2018
Words like machine learning, big data, data science … are in everyone’s mouth. In fact, the profession of data scientist has been described as the sexiest of the 21st century. Many people talk about the data revolution and artificial intelligence, but what is machine learning really and why is it so famous?
I will try to shed some light on this.
Tom Mitchell defines machine learning in one of his books as: “The study of computer algorithms that automatically improve their performance thanks to experience. It is said that a computer program learns about a set of tasks, thanks to experience and using a measure of performance, if their performance in these tasks improves with experience. “
That is, algorithms that learn and self-improve “alone” thanks to experience. The fact that they do it alone is in quotation marks because they do it using data, past experiences. Unlike models in which a business expert assigns rules and models something according to their knowledge (past experience), statistical models and machine learning models allow data to speak and obtain relationships automatically.
Another definition is given by Drew Conway, who with the following Venn diagram defined machine learning as the union of hacking skills and mathematics & statistics.
But why is there so much talk about it now?
A lot of the methods used in machine learning and statistical modeling have been with us for several decades. Algorithms such as neural networks or vector support machines (SVMs) were devised quite some time ago, even some of them fell into disuse.
Some of the main reasons for the current Boom of these techniques are:
- On the one hand, the computational capacity of computers has been increasing and it is now possible to deal with problems that could not be dealt with before. This increase has been vertical (improvement of individual computing capacity, CUDAs …) and also horizontal (increase in computing capacity when working with several computers at the same time using the Big Data paradigm).
- On the other hand, the revolution of big data, motivated by digitization, has meant a huge increase in data that can be processed and modeled to obtain knowledge of them. Years ago there was much less data, being possible to see statistical models of a few hundred records.
We are currently living in an exciting period in which data and the application of techniques that extract value from them will be strategic for many countries and sectors. You only have to look at the investment that China, the US and other countries are making to realize that Machine Learning and current modeling techniques are the present and the future.