Neural networks and mobile applications
One of the top managers of Google Sandar Pichai said in 2014 that the main priority of the company at this stage of its development is machine learning. Two years later, after this statement, in March 2016, the Alphago program beat the best player in Professional 9 Dana Li Sedol. This program developed by the Deepmind startup and today belonging to Google works on the algorithm of self-learning neural network. After the defeasible victory of artificial intelligence over the human genius began a boom of neural systems.
What is the neural networks where they are applied and why today they are mentioned so often, we will tell in this article.
Navigation
What is neural networks and what tasks they solve?
Humanity has long begins to develop in the field of artificial intelligence. Neural networks, one of these developments. The principles of building such networks lies in the "copying" of the human nervous system. Their main feature is that they are capable of self-study. That is, they can work on the basis of the data obtained earlier.
But, neural network is not only imitation of human systems. They also resemble the nervous system. That is, consist of separate computing elements. Which can be compared with human neurons. All data that the network receives is consistently undergoing the processing of several neural segments (layers), which consist of several "neurons".
Since the networks are known for the parameters of each "neuron" and the entire network as a whole, it may, as a result of previously obtained data to change its work. That is, "self-study."
Today, neural networks are used quite often. For example, many large online shopping use them in order to offer their customers more suitable for them products. Mathematical models for recognizing and playing speech, as well as recognition and image processing are also widely used.
The boom on unmanned cars also opened a new scope of use of neural networks. They are used to lay the route and the operation of navigation systems.
The manufacturers of antivirus software also drew on them. Such developments in the field of artificial intelligence allow you to protect information from cybercriminals and identify illegal content on the Internet.
Already, futurologists "draw" an amazing picture of the future, where the scale of neural networks will increase repeatedly. For example, robots will appear, gardeners. Which thanks to hundreds of built-in microcamers will receive data, analyze them and separate weeds from cultivated plants. Thanks to "self-education" such robots are capable of applying individual processing to each plant.
In the Internet developed now, the neural networks will be able to create voice interfaces for "communication" of various objects of such "web". Algorithms of such a mathematical model will be able to replace a person in call centers, controllers in warehouses, translators, consultants and representatives of other professions. According to many analysts, this neural networks will need only 3-5 years.
Already today, neural networks allow you to take some solutions for a person. Let them be called too intellectual yet, but the victory of the program based on such algorithms allows you to believe in their success in the future.
Why are neural networks have become so popular now?
About artificial intelligence Humanity thought for a long time. On the use of neural networks to solve this issue, 70 years ago. But, only today it has become possible in those volumes that are necessary.
The popularity of neural network today is due to the increased speed of the "neurons". That is, individual computing elements. Thanks to which it was possible to reduce the time for the stage of "training." Tens of millions of data required for the operation of such networks due to modern processors and video cards are processed quite quickly. Modern technologies of accelerated learning and led them to advanced artificial intelligence.
Algorithms of neural systems in mobile applications
Recently, there have been a lot of entertainment projects on the mobile application market, which use neural networks. Of course, the main advertisement made them the application Prisma. In addition, it is worth noting a competitor of this MLVCH application, as well as a popular MSQRD video service. Which also uses the described mathematical algorithms.
Prisma.
The PRISMA application, Russian developers, transforming a photo of the pictures of famous artists, perhaps the most famous application of neural networks in modern life. First of all, Prisma demonstrates the capabilities of such mathematical algorithms when developing mobile applications. But, such stylization of images can be applied not only in entertainment purposes, but also in design, animation, computer graphics and other similar regions.
Yes, the creators of Prisma cannot be called pioneers. Using a neural network for image processing was used before. For example, similar algorithms were used to create a film in Van Gogh style, but the processing of one frame took about 3 minutes. Modern technologies allow it much faster. In the same prisma, the photo processing takes a little more than 1 second. And this is despite the fact that the photo is loaded to the server, processed and sent to the user.
When processing photography, objects displayed on it are defined. After that, the style of the chosen artist is applied to the picture. That is, the application with the help of artificial neural networks "duplicates" the work of the artist's brain. It is amazing, fascinating and very beautiful.
Now in the "prism" you can choose filters imitating styles 21 artists - from Edward Mink and Mark Stegal to Van Gogh and Picasso. According to the developers, the applications algorithms are already soon helping to choose suitable filters in automatic mode.
MLVCH
MLVCH is the main competitor Prisma.
MLVCH is the main competitor Prisma. It works on a similar algorithm, but is distinguished by a more complicated elaboration. If the PRISMA image is processed in 20-30 iterations, then in MLVCH for 100. Which allows you to get a more detailed image at the output. In addition, MLVCH is 2.5 times more filters than in PRISMA.
As for the minuses, this complexity of photo processing in MLVCH increases the time of receiving the finished option. In addition, only one photo can be processed free of charge on a day. For all subsequent will have to pay 75 rubles. In addition, you can buy a package of 100 photos for 2990 rubles.
Where else do neural networks apply?
Of course, applied products using neural networks are most often used to analyze images. But using the Clarifai application, you can go further and recognize the image on the video. Another similar service offers the Russian company NTechLab. She developed the FinDFACE application. With which you can by a photo of the user to find his profile in the social network "VKontakte"
Service "Auto.ru", which today belongs to Yandex, on the photo independently recognizes the brand and model of the depicted car. The principle of "computer vision" used in such recognition also works on neural network technology.
Using the Ostagram application, you can not only bring styles, but even redraw pictures. And with the help of Deepomatic, the user based on what pictures he distributed on the Internet is offered products from online stores.
Of course, the analysis of pictures with advertising and entertainment targets The area of \u200b\u200bapplication of a neural network is not limited. AVA startup allows you to follow your health. The application analyzes photographs of dishes and food and issues information about calories and various additives.
Not so long ago, with the help of the Yandex neural network algorithm recorded the music album "Neural Defense". The project that was called "AutopoET" generated stamped texts.
Uses the described mathematical algorithms and another leading Internet search engine Google. In November 2015, this company introduced a special function to his Gmail mail, thanks to which the service itself generates three answers to an incoming letter. You can choose the best mouse with one click.
Also neural networks are used when creating unmanned cars and flying devices. First, the person is managed by these means of movement. The data are collected in a special base, which is used for "learning" autopilot. Already in Switzerland there is an unmanned aerial facility that uses the principles of neural network and is used to search for lost tourists.
Unmanned car
Popular recent bots that are used in social networks and messages can also use the described mathematical algorithms. The American launch of LUKA successfully uses a neural network for learning his bots. And using the Eterni.Me application, you can create an avatar and "teach" it to be your own owner. For this neural network collect information about the owner of the avatar and introduce it to a virtual object. Such an object after the death of his prototype will be able to continue the "life" of its owner.
In addition, this bot avatar can be created to communicate with people who do not want to spend time. Burner successfully tests such a product. We are waiting for the adoption of a special application to create your bot.
Also neural networks in the future will be used to form a medical diagnosis or "reading" results of analyzes. Companies such as AtomWise, Enlitic, Quantified Skin, etc. already work in this direction. And the DEEP Genomics project can analyze the genome and predict the development of possible diseases.