What is a neural network? How to train it and why is it necessary in 2021?

Over the past few years, artificial intelligence has quietly shaken off the “sci-fi” and “game design” tags and made its way into daily news feeds. Something mysteriously called a “neural network” identifies people from photos, drives cars, plays poker and makes scientific discoveries. However, from the information we know, it is not always clear what these mysterious neural networks are: complex programs, special computers, or racks with rows of servers?

Of course, one can guess just from the name that in neural networks developers have tried to copy the structure of the human brain: as we know, it consists of many simple cell-neurons, which exchange electrical signals with each other. If you compare a neural network to a child, in simple terms, it is the same child with “zero” memory, which is ready to learn all the information it is taught. Learning comes down to a person solving first elementary tasks, and then more complicated ones. All data is loaded into the neural network memory. Over time this data is accumulated, and now the neural network can not only distinguish red from blue but also build a model of a suspension bridge, for example.

These days, we are increasingly faced with exploring new planets, complex heart and brain surgeries, the need to develop vaccines for new viruses, etc. It is inherent in human beings to make mistakes, but none of these areas gives you the right to be wrong, because the most important thing at stake is the life and future of all mankind.