What is machine learning in simple terms?

A person differs from a computer in that he learns from his mistakes and actions. Machines need to be told what to do, as they obey strict logic and make no sense. So we write programs that give the silicon intelligence precise instructions. Machine learning simply forces the computer to draw up detailed, step-by-step instructions on its own from past data skills for admin.

What is machine learning in simple words: algorithms, methods, what you need to know

What is machine learning?

In fact, these are computer programs with certain algorithms of action:

the process begins with the analysis of an array of information,

then a pattern or pattern is revealed,

after all, direct experience is formed

and instructions are drawn up based on it.

The system, without human intervention, is capable of offering turnkey solutions based on real data.

Machine learning methods

It is customary to distinguish machine learning methods (algorithms, that is, their main learning strategy) by the degree of human control and intervention. Today we are dealing with both minor applications that help speed up routine processes, and with fully automated complexes for non-stop studying of information arrays at large enterprises.

Supervised machine learning algorithms

Controlled "learning with a teacher» ( Supervised the Learning ) involves the use of AIs in all studied in the past to new data. User-marked examples are used to predict future events. The method is based on the analysis of the set of training data designated by the "teacher". A machine learning algorithm creates a supervised predictive function of the output values.

Unsupervised machine learning algorithms

"Learning without a teacher» ( unsupervised the Learning ) is used when the information is classified and no one not marked. Unsupervised learning analyzes how systems can infer a function to describe a hidden structure from untagged data.

Partial machine learning algorithms

Semi-supervised learning algorithms fall somewhere between supervised and unsupervised learning. They use all available data. Usually a small amount of information is labeled "teacher". This helps to clarify the actions of the machine when working with a large amount of untagged data and to speed up the process. Systems can significantly improve the accuracy of work using this method.

Reinforcement learning algorithms

In the role of "reinforcements" ( Reinforcement the Learning ) serves traditional human approach method of "trial and error". The machine interacts with its environment, performing actions and detecting errors or rewards (achieving goals). Reinforcement learning algorithms allow software agents to automatically determine the ideal behavior in a specific context to maximize its performance.