The amazing capabilities of neural networks in 2019

It seems that not a day goes by without a message with the words “artificial intelligence”, “neural network”, “machine learning” in the news. This is not surprising, the algorithms are constantly being improved, they gain new knowledge, including information about each of us. And this draws out very interesting prospects for the future, with personalized goods, food and entertainment. But first things first.

What surprised neural networks?

Some news about neural networks that appeared in recent months a couple of years ago could be easily sent to the shelf of fantastic stories. But no, it's not fiction, it's 2019.
')

Neural networks understand what babies want

Researchers from New Jersey have developed a neural network that can distinguish babies' cries from each other and classify them. Tests with more than a hundred (to date) babies from 3 to 6 months showed that in the absolute majority of cases the neural network correctly understands what the child wants: to eat, sleep, change the diaper, attention, feel pain or other discomfort.

Neural networks “liven up” images

Employees of Samsung AI Center-Moscow and specialists from Skolkovo created a system capable of creating an animation from just a few (from 1 to 8) human images (photos or even portraits). As a result, we can look at the rather real moving faces of Albert Einstein, Merlin Monroe, Fyodor Dostoevsky, and many others. Looks impressive!

Neural networks create photographic images.

This year, several neural networks were presented, capable of quickly drawn sketches by a person to turn into real-looking images (for example, a small house in the form of a square and a triangle or a drawn smile on a person's photo).





Moreover, neural networks appeared independently creating indistinguishable faces from real people, images of animals, transport, residential premises and much more.

Neural networks re-create portraits of people

Massachusetts Institute of Technology continues to amaze. The Speech2Face neural network represented by this institution creates portraits of people only after hearing a sample of their voice. And just look at how accurately in most cases the algorithm shows the gender, nationality and age of people. And in some cases the faces themselves are quite clearly similar to the original.

Neural networks recognize age by eye movement.

And at the University of Minnesota, a neural network was developed, which could only correctly determine the age in 83% of cases by the trajectory of eye movement of children. True, only children under 3 years old participated in the research.

Neural networks write "human" texts

Earlier this year, the OpenAI company announced the creation of a neural network capable of writing a text (whether it be news stories or a whole story) indistinguishable from human articles and works. Fearing inappropriate use of the program to create false news notes, the company did not present this neural network to the general public.

Neural networks create false movements of people on video

The developers at Facebook AI Research have taught their neural networks to recognize a moving person in a video, replace everything in a video except a recognized object, and even add new movements to a person in a video. Moreover, the “captured” human video image of a neural network allows you to control it using a keyboard, as in a computer game.

Neural networks cancel photoshop

The nightmare of many became a reality - a neural network appeared that was able to see if the image was processed in Adobe Photoshop, and then recreate the original image. So far, the program defines only one (albeit, the most popular) tool “Facial Plastic”, but the creators of the neural network believe that in the near future no image editing will go unnoticed.

Neural networks will decide what to offer you to eat.

McDonald's acquired Dynamic Yield, which develops neural networks for personalized advertising. So, perhaps, soon, you will be offered food, the acquisition of which is very likely you will be very difficult to refuse.

Neural networks coming up with new sports

You are not a fan of any particular sport? Perhaps neural networks in the near future will be able to come up with a new sport that you enjoy. So, AKQA , using its neural networks, based on 7,300 rules from 400 other sports, was able to create a new Speedgate sport. The main rules of this game can be found in the video, which also shows how the creators of the neural network play Speedgate.

Neural networks make music

It is not the first year that there have been reports that neural networks are writing music in one or another style. Earlier this year, Yandex introduced a neural network that was able to write a play for a symphony orchestra with viola. Don't like the classics? Then, perhaps, you will enjoy the Dadabots neural network, which live on YouTube channel non-stop composes works in the style of death metal.

Neural networks create alcohol

Developers from Fourkind and Microsoft have developed a neural network that has created a new sort of whiskey for the Mackmyra distillery in Sweden. At the same time, more than a hundred parameters were taken into account - from the ingredients and production methods to the characteristics of infusion of the resulting drink.

What about personalization?

A significant part of a person’s life goes on the Internet, in various social networks. Not surprisingly, the algorithms from Facebook, Instagram, YouTube, Google, Amazon, Twitter, etc., know many people almost better than they do. And this direction will only develop. Is it bad? It is difficult to say for sure.

But it is quite possible that the future can await us, where our own music will be written for each of us, to create pictures, to write stories, even to be offered certain foods and unique drinks. The same news for different people will be heard in a different way. One person will see dry statistics in two lines, the other - a colorful description of a process in the form of a longrid, all depending on the preferences of each.

By the way, even movies and TV shows can also become personalized, with the help of adding magic technologies. For example, Netflix is ​​already experimenting with scenes that depend on the decisions and actions of the viewer. And if you suddenly want to see other actors playing roles in the film? Neural networks are already beginning to come to the rescue. DeepFake technologies are getting better every year and soon you can replace the faces, bodies, clothes of any personalities in a video in a few minutes.



For example, you can take a look at the scene from “Terminator 2: Judgment Day” in which Arnold Schwarzenegger was replaced by Sylvester Stallone.

Summary

And this is not all the events associated with neural networks, only this year. Given the above news, it is not surprising that the governments of many countries are increasingly paying attention to issues related to artificial intelligence. So, on June 8 of this year, representatives of the G-20 countries for the first time signed a document containing principles for working with artificial intelligence.

One can confidently assert that this sphere will develop even more rapidly, since no one wants to remain behind the progress train that is gaining momentum. For example, in Russia, at a meeting with the president on the development of technologies in the field of artificial intelligence, it was said:

... to develop solutions that can provide superiority [of artificial intelligence] over man for special tasks. And by 2030, we must ensure the superiority of man in a wide range of tasks.