Machine Learning in Kazan, or how machine learning specialists are trained in Tatarstan

Tatarstan has long shown ambitiousness in the development of high technologies. Recently, Kazan IT Park celebrated its 10th anniversary - the very one where startups and small companies are torn to an international level. The city of Innopolis also demonstrates power despite skeptical forecasts: according to statistics, in 2019 the number of residents increased by a third, and new offices of world corporations opened in the technopark. Okay, how are things going with global IT trends — machine learning and artificial intelligence technologies?

The main pioneer of this area in Kazan is Evgeny Razinkov, Ph.D. (Physics and Mathematics), lecturer at Kazan Federal University, head of the machine learning and computer vision department of the FIX Group of Companies and director of science at Pr3vision. Once he was an ordinary student of the VMK, and a year ago he launched a master's program in Machine Learning and Computer Vision in his native university. Evgeny takes the most talented colleagues and students to his team, which is engaged in scientific development and collaborates with large IT companies. They found out everything about Machine Learning in Kazan from him.



- At what point did you realize that it was time to develop the direction of machine learning in Kazan?

- Everything happened somehow naturally. In 2014, he became interested in machine learning himself, and in May of that year he met the Czech professor Jiri Matos, who became my scientific adviser. In December 2014, he went to his internship in Prague for the first time, and in February he began to conduct seminars on machine learning in KFU. It was something like this: I come to my first seminar, collect students and say: “Guys, now there will be a seminar on machine learning. I don’t understand much myself yet, so let's sort it out together. ”

During the semester, I read these seminars and studied with students, and then it turned into a curriculum, which was included in the curriculum and was already taught to bachelors as an obligatory subject. I independently developed and started teaching a course on computer vision. Read Data Mining. Then he developed his course on Deep Learning. I think this was the first such course that at that time could be found in Russia. For example, in the Czech Republic it is being developed only now, and I am even a little proud that I have been reading it for a long time. And a couple of years ago I noticed that students began to show more interest in my lectures. They come to me not on schedule, asking how to get to my classes.



- What platforms are now preparing specialists of the Ministry of Defense in Kazan, and which of them do you consider the most effective?

- Disclaimer: here is the story about my obvious conflict of interest :) Of course, first of all I will name KFU, Institute of Computational Mathematics and Information Technologies (former faculty of the VMK), department of system analysis.

This will sound immodest, but I really believe that we train really cool specialists. Now we have seven teachers - all the guys from my team, and most have at least 3 years of experience in this area. According to my information, there are almost no experts in such experience in Kazan except us. I give lectures on machine learning in the 2nd and 3rd year of undergraduate studies, in the fourth year of undergraduate studies and in the master's program - computer vision. I also have a master course in deep education for undergraduates.

Plus, in our department of systems analysis in 2019, a new master's profile appeared - “Machine Learning and Computer Vision”, which I opened with my team. In this master's program I give courses in deep education, computer vision, in the next semester I will give lectures on teaching with reinforcement, natural language processing. My colleague Ruslan Nigmatullin gives lectures on digital image processing and digital signal processing. In total, we have 9 lecture and practical courses related to artificial intelligence. The guys who once wrote my diploma theses are now conducting practical classes in our graduate school. Last year we got a good competition for this profile, which makes me very happy.

My lectures are open to all comers (those who have access to KFU buildings). Now I am posting edited videos with lectures on the YouTube channel. In the near future I plan to stream lectures online in connection with the COVID-19 pandemic - all this will be in the public domain.

As far as I know, also in Kazan such specialists are trained at KNITU-KAI and Innopolis University.



- Where are your graduates going to work? They want to develop in Kazan or are they striving for Moscow, abroad?

- It seems to me that such a small number of graduates - strong guys will have enough work here. Kazan is developing quite dynamically in the IT sector. Some of my students are leaving for Moscow. In general, of course, good specialists in this direction are expected everywhere.

- You manage the department of machine learning and computer vision in a large holding. How do you rate the role of such a department across the company? Are such companies considered steeper / more prestigious / more promising than others?

- Now the topic of artificial intelligence and machine learning is on the rise, most of the innovations in the world are associated with them. It’s hard for me to think differently, but in my opinion, these technologies are super important for every company that plans to develop an innovative business. Because actual business innovations are now very often tied to our algorithms, and it is difficult to manage such companies without machine learning departments.

Even if the corporation is large in appearance, but it does not have expertise in machine learning, computer vision, and AI, there is a feeling that such a business is a thing of the past. It is likely that several years will pass and they will lose to their smaller competitors, who now pay more attention to our technologies and algorithms.

- What do you consider the greatest achievement of your team?

“We have two great reasons for pride.” The first is that we really successfully close many projects. Most of our developments are commercially implemented and delight our customers. Many projects are protected by the NDA and are a trade secret of the companies with which we cooperate, so there will be few specifics.

Of the open projects that we implemented - a tool for the cashback service Backit of the FIX Group of Companies in Kazan. We have created a system that allows you to find the right product with a cashback from a photograph. Let's say a user wants to buy a specific sneaker model with a cashback. He uploads a photo of the product into the service application, and a specially trained tool gives him a link to this purchase with a cashback. Now, more than 1000 online stores with a wide variety of assortments, from cosmetics to building materials, are connected to the Backit cashback service. Our algorithm searches for clothes and accessories.

Our second achievement is the quality training of specialists, which we have fully built up. Many people from my team teach students, these guys write a diploma with us, then a master's thesis. And already in the middle of the magistracy, we have enough strong guys who then come to work with us in commercial companies.



- What principles, in your opinion, are most important in training specialists in this area?

“You can talk a lot about math here.” I believe that a person who does not have a good mathematical background is very limited in his development and cannot claim rapid development in machine learning.

Another important point. When machine learning specialists design a system, accuracy is always measured at the end. The question often arises - are these measurement methods clear to the customer? More often than not. The metrics that we use are complex, and people without special technical education may be incomprehensible. Businesses are usually interested in KPIs that are tied to business tasks, and machine learning works with completely different metrics. This makes our specialists responsible - to be able to switch to the customer’s language and correctly convey the results of the work.

The question of professional ethics immediately arises. You can easily keep the customer in error and “delight” with beautiful numbers, which will have a rather weak connection with the use of the system in real conditions - unfortunately, some experts do so. Once we worked with a customer who talked about what other specialists were offering him. Competitors promised him certain indicators of accuracy, but did not explain how exactly this metric is calculated. This approach does not seem ethical from a professional point of view.

Therefore, an important principle in the training of machine learning specialists is to pump their willingness to speak the customer’s language. Honestly translate mathematical metrics into business logic, into business language. And you should be the person who, if something happens, can honestly tell the customer: "Machine learning is not applicable here, do not hire me."

And there is such a thing - the Dunning-Krueger effect. This is a psychological phenomenon when people with low qualifications consider themselves cool specialists, and really cool specialists consider themselves not very cool. If you google the Dunning-Krueger chart, everything becomes clear.

Machine learning is now such a hype field that the temptation to consider yourself a good specialist is great. Especially when online courses promise to "become a professional in three months, or even a month." This is compounded by the fact that many companies do not yet know how to objectively assess the level of training of a specialist in this area. If the company did not have such competence and is just being created, candidates for this place may try to sell themselves in any way. And here again the question of honesty is relevant.

No matter how much a specialist would like to get a job in this field, he should not promise the customer something that he will not be able to handle, even if the customer "is glad to be deceived." This is very harmful for the region - customers have high expectations. What does this lead to in terms of technology development? Moreover, in one day the belief in our methods and algorithms is weakening.

It seems to the customer that with the help of machine learning, he will immediately begin to do cool things, hire specialists who, let's say, are unjustifiably optimistic. As a result, all this leads to nothing, and the customer says: “I was disappointed in the Moscow Region. This does not work".

For me, as a specialist who makes money on this, this is a big problem and pain. I would like other machine learning experts to be responsible for what we do. So that everyone soberly evaluates their skills and how much the problem is solved, he knows how to return the customer "to the ground." So that there is no disappointment in technology and everyone can work normally, productively.

- When will the Kazan community of machine learning specialists glorify Kazan all over the world?

- Here you can probably only laugh it off :) In general, what glorifies the whole world? The situation when a company has a lot of money and it is ready to invest in third-party things - not in what it makes money. And when it comes to what she earns on - as a rule, this is a trade secret, and you do not glorify Kazan all over the world.

Yes, there are such companies as DeepMind, OpenAI - there is a lot of money, people teach bots to play warcraft, DotA. There seems to be no direct application, but there is hype, because everyone is interested. Such things are glorified all over the world. Still, of course, scientific achievements have such an effect. There are universities such as Oxford, Stanford, MIT - they also have such resources.

The main thing that we can become famous for is educational activity. The greatest that we can count on is to become known throughout Russia as a center for the training of cool specialists. So that one day people will notice our work and say: “Oh, Kazan has a strong ML school!” Talented students from neighboring cities would draw in to us, together we would create even more interesting things.

So a reasonable ceiling is popularity as a strong machine learning school in Russia. I can’t imagine it on a global scale. Probably, the situation in higher education and commercial companies is not the same. Perhaps when we gain a greater advantage in the commercial plan, they will begin to invest more readily in our technologies. Now the business is waiting for a quick effect, but in machine learning a different approach is needed, more fundamental and strategic - only in this case something good can happen. Let it be for now.

---

Contacts Eugene:

Telegram channel (announcements of lectures and seminars, important articles on machine learning): t.me/razinkov_ai

YouTube channel (video of lectures): video.razinkov.ai

Public in VK (schedule, announcements of seminars): vk.com/razinkov_ai