Machine Learning: Theory and Practice

Practice makes perfect. This popular sentence needs no explanation and can be applied to every aspect of our lives. Our brain, as well as our muscles, develop through training and exercise. Experience helps us do our job better. But can we apply this method to a machine? Can a computer become better through practice?

What is Machine Learning?

Machine learning is the field of computer science focused on research of algorithms and systems, that can learn and improve from experience. By experience, we mean information coming from data processed by the computer. On the basis of these data, the system can improve i.e., increase its capacity to solve given task.

One of the first examples of such machine was a checkers playing program designed by IBM engineer Arthur Samuel. This program was able to remember previously played games, and base its decisions on that “memory.”

Most of the created software operate by pre-defined code, within the frames specified by its programmer e.g. built-in checkers winning strategy. A self-learning machine, however, has only a set of rules and winning conditions, so it has to learn by practice to be able to win. Above applies not only to games, but also for more advanced applications like pricing, categorizations or different predictions.

How Does Machine Learning Work?

Through the learning process, algorithms can detect specific patterns and dependencies to build a form of knowledge. We can distinguish two basic types of machine learning.

The first one is supervised learning. In this model, the machine is trained using already labeled data. A computer is given examples consisting of the input object and corresponding output value. The task to be performed by the computer is to predict the output from the input data. A simple example would be the categorization of images. Let’s say that you give the system photos of cats and dogs with appropriate tags on every photo. The learning algorithm should process that data and be able to assign later added photos, which have no labels, to proper category — cat or dog.

There is also a second possibility, which is a bit more complicated. It is called unsupervised learning. In this situation, we give the system only an input data, without the desired answers. The software has to find the patterns and relationships independently. A typical example of unsupervised learning is the task of detecting certain regularities in the output data, and grouping them on this basis in certain categories. This most common method of unsupervised learning is called clustering because the input elements are grouped in a relatively homogeneous classes (clusters).

Machine Learning in Practice

There are many possibilities for practical use of machine learning. We even use them unknowingly almost every day.  The appearance of our Facebook news feed is largely based on machine learning.  Each time we click the like button or stop scrolling to read some interesting story, Facebook algorithms learn something about our interests and behavior. The content you see is influenced by your connections and activity. The goal is to show you the most interesting news from friends you interact with the most.

Machine learning can also be extremely useful in any field that requires analysis of big amounts of data. It is a very powerful tool in economics, making possible to anticipate trends on financial markets. In medicine, machine learning can help diagnose patients in time to prevent the further progress of a disease. Automated data analysis is also a top priority for NASA, which wants to use it for systematizing astronomical objects and building autonomous spacecraft.

Of course, in some situations, we will not be able to provide information in the form of computer data. An experienced doctor will be able to take into consideration some factors, that may be omitted by a machine. However, when we are dealing with complex and a large amount of data, computer capabilities are unparalleled.

Machine learning is still developing and finding new practical uses. The number of possible applications is extremely vast, and we can already say that in the future every aspect of the technology will include some implementation of machine learning algorithms.