Machine Learning: Potential and Opportunities

/ photo Jaro Larnos CC

Cloud technologies are widely used in various scientific fields: once we talked about how clouds are used in physics and astronomy , as well as geography and genetics . Virtual infrastructures allow scientists to process a wealth of information in the shortest possible time, leading to new discoveries.
')
But there is another technology that can change our understanding of information processing. We are talking about machine learning, which has recently gained particular popularity.

A few years ago, Google completely redid its image search and speech recognition services, introducing machine learning elements, and on June 16 of this year, the giant announced the expansion of a research center in Zurich, which will be engaged in developments in the field of AI, processing of natural language and machine perception. This means that Google is going to develop systems that are able to think, listen and see.

Senior Researcher Greg Corrado (Greg Corrado) argues that the active introduction of machine learning can bring no less benefit than the creation of the Internet. This may lead to the fact that we will not need to understand all the details of certain processes, it will be enough to load data into the system, on the basis of which it will start learning.

The most promising direction in machine learning is the so-called deep learning. It is built on neural networks (NN), which require a large amount of data to learn. The NAs were first described in the 1930s, but they were actively used only in the last 3-4 years, as the power of computers increased dramatically.

Last year, Google launched its open library for deep machine learning TensorFlow . So the company is trying to draw attention to the project and develop it by third-party developers. The system presents the calculations in the form of a data flow graph. Its main feature , unlike other platforms like Theano and Torch, is support for distributed computing.

The company TensorFlow use almost in all directions: from speech recognition to searching for photos, but in fact it will be more useful for scientists conducting experiments on deep learning of neural networks, as well as companies that need to quickly train and test their models. Feel TensorFlow do it yourself by clicking on this link.

AI goes to writers

The journalist of the Guardian Alex Hern (Alex Hern) in his article told about his attempt to teach the simplest recurrent NA so that she could logically complete the sentences. As a training data, he took 119 MB of text from The Guardian articles. On other interesting options for the use of recurrent NA, read in this article.

Half an hour after starting the learning process, Alex saw that progress was only 1%. He realized that the power of his computer is not enough and decided to rent a server in the cloud. This allowed to complete the learning process for 8 hours.

It turned out, to put it mildly, not very cool. The computer had to continue the following phrase: “The fateful decision to stay in the EU, taken on Thursday, was ...”. As a result, the system offered options such as "... based on a promise made in several statements" and "... a member of the 2015 opposition party". On the one hand, complete nonsense, on the other - there is a positive point: if the car learned to write articles for The Guardian, Alex and his colleagues would be left without work.

This result is quite understandable. The neural network used in training could only recognize characters: she did not know what a word was and did not understand grammar. Moreover, in order for the network to adequately compose sentences based on real-world data, it needs to transfer a much larger amount of data for training. A set of articles of one edition is not enough. The Atlantic journalist and a writer who previously worked on Twitter also wrote about such experiments.

Humanity is in a hurry to help

AlphaGo, an AI-based program that recently beat the world go champion, is one of the most prominent examples of deep learning. The program involves two types of training: training with a teacher, when data from all matches played between people are used, and reinforcement training, which means that the program plays against itself and learns from its mistakes. But with all this, as it turned out, some things AlphaGo just can not learn on their own.

According to the leader of the research team DeepMind, which developed the program, the system was well aware of which areas of the playing field it should focus on. However, the program does not know when it should stop the “thought process” and make its move. This is an important moment in the game, since in professional matches there is a complex time control system. For example, in the game against world champion Lee Sedol, the players had two hours to think over the moves, as well as three minutes of extra time, which were added if none of the previous moves were completed in less than three minutes.

The developers did not add time control rules to the program, but only introduced a restriction by developing a special algorithm. Later it was optimized by a program based on a series of experiments, but the fact is that without the help of a man, AlphaGo would not have been able to beat the champion.

This situation with AlphaGo leads us to the idea that the progress of AI training can be accelerated if ordinary users are involved in learning systems. For example, the popular computer game Minecraft is now becoming a platform for human and machine collaboration.



Recently posted on GitHub Project Malmo, launched by Microsoft, is a platform for exploring the capabilities of artificial intelligence. The task is to train the character of the game to perform various actions, starting with a bridge crossing and ending with the construction of complex objects. In addition, the project allows you to organize a joint game of AI with a person, as well as communication between them using a special chat.

According to project manager Katja Hofmann (Katja Hofmann), the goal of Project Malmo is to create an AI that will learn from users and help them solve their problems. The program uses reinforced learning algorithms. For example, you can teach a car to navigate in a room with many obstacles. Regular players can give hints or instructions that the AI ​​will gradually learn to recognize and make the right decisions based on them.



The Minecraft platform has also been used to train a robot at Brown University. According to one of the university professors, Project Malmo will be an effective method of collecting data on human interaction with AI. Perhaps soon we will be able to fully communicate with artificial intelligence.

PS Additional materials on machine learning

• Courses on machine learning from Coursera and Udacity ;
• A deep learning course from Udacity ;
• Video lectures of Professor K.V. Vorontsov on machine learning;
• TensorFlow mini tutorials for beginners;
• Useful article by graduate student Andrei Karpaty (Andrej Karpathy) from Stanford on the use of recurrent neural networks.

PPS Here are some more interesting materials from our blog:

• Foreign or local data placement in the cloud
• Storage technologies in the insurance industry
• Cloud technologies in the tourism business
• How to increase the level of cloud infrastructure security