AI, student and big prize: how to do machine learning in 8th grade

    Hello, Habr!

    We want to talk about such an unusual form of earnings for adolescents as participation in hackathons. This is both financially profitable and allows you to put into practice the knowledge gained at school and through reading smart books.

    A simple example is last year’s hackathon of the Academy of Artificial Intelligence for students. Its participants had to predict the outcome of the game Dota 2. Then the winner of the competition was Alexander Mamaev, a tenth grader from Chelyabinsk. His algorithm most accurately determined the team winner of the fight. Thanks to this, Alexander received solid prize money - 100 thousand rubles.




    How Alexander Mamaev ordered the prizes, what knowledge the student lacks to work with ML, and what direction in the field of AI he considers the most interesting - the student said in an interview.

    - Tell me about yourself, how did you get involved in AI? Was it difficult to enter the topic?
    - I am 17 years old, this year I am finishing school, and recently I moved from Chelyabinsk to Dolgoprudny, this is near Moscow. I study at the Kapitsa Physical-Technical Lyceum, this is one of the best schools in the Moscow region. I could rent an apartment, but I live in a boarding school, it’s better and easier to communicate with people from the lyceum.

    For the first time, I heard about AI and ML in the year 2016, probably, when Prisma appeared. Then I was in 8th grade and was engaged in Olympiad programming, attended some olympiads and found out that ML meetings are held in our city. I was interested in sorting this out, understanding how it works, and I started going there. There he first learned the basics, then he began to study it on the Internet, at different courses.

    At first, there was only a course in Russian from Konstantin Vorontsov, and the manner of teaching it was tough: it contains many terms, and there are many formulas in the descriptions. For the eighth grader, it was very difficult, but right now, thanks to the fact that I went through such a school in the beginning, the terms do not present difficulties for me in practice in real tasks.

    - How much do you need to know math to work with AI? Is there enough knowledge from the school curriculum?
    - In many ways, ML is based on the basic concepts of a school of grades 10-11, basic linear algebra and differentiation. If we are talking about production, about technical problems, then in many ways mathematics is not needed there, many problems are solved trite by trial and error. But if we talk about research, when new technologies are created, then without mathematics there is nowhere. Mathematics is needed at a basic level, if only to know how to make a matrix application or, relatively speaking, count derivatives. There is no escape from mathematics.

    - In your opinion, can any student with a natural-analytical mindset solve problems in ML?
    - Yes. If a person knows what lies at the heart of ML, if he knows how the data is arranged and understands basic tricks or hacks, he will not need a matan, because many tools for work have already been written by other people. It all comes down to finding patterns. But everything, of course, depends on the task.

    - What is the most difficult thing in solving ML problems and cases?
    - Each new task is something new. If the task already existed earlier in the same form, it would not have to be solved. There is no universal algorithm. There is a huge community of people who train their skills in solving problems, tell how they solved problems, and describe the stories of their victories. And it’s very interesting to follow their logic, their ideas.

    - What cases and tasks are you most interested in solving?
    - I specialize in computer linguistics, I am interested in texts, classification tasks, chat bots and more.

    - Do you often participate in AI hackathons?
    - Hackathons are, in fact, another system of olympiads. The Olympiad has a set of closed tasks, with well-known answers that the participant must guess. But there are people who are not strong in closed tasks, but tearing everyone open. So you can test knowledge in different ways. In open-source tasks, technologies are sometimes created from scratch, products are developed quickly, and even the organizers often don’t know the right answer. We often participate in hackathons, so you can earn money. It is interesting.

    - How much can you earn on this? And how do you spend the prize money?
    - My friend and I participated in the VKontakte hackathon, where we made an application for searching for paintings in the Hermitage. A set of emoji, emoticons was displayed on the phone screen, you had to find a picture using this set, the phone was pointed at the picture, it was recognized with the help of neural networks and, if the answer was correct, points were awarded. We were pleased and interesting that we managed to make an application that allowed us to recognize the picture on a mobile device. We went first in first place, but because of the legal formality we flew past a prize of 500 thousand rubles. It's a shame, but this is not the main thing.

    In addition, he participated in the competitions of Sberbank Data Science Journey, where he took 5th place and earned 200 thousand rubles. They paid a million for the first, and 500 thousand for the second. Prize pools are different, now they are increasing. Being in the top, you can get 100 to 500 thousand. I put aside prize money for training, this is my contribution to the future, the money that I spend in everyday life, I earn myself.

    - What is more interesting - individual or team hackathons?
    - If we are talking about the development of a product, then this should be a team, one person will not be able to do it. He will get tired, he needs support. But if we are talking, for example, about the hackathon of the Academy of AI, then the task there is limited, there is no need to create a product. There is interest in something else - to overtake another person who is also developing in this area.

    - How do you plan to develop further? How do you see your career?
    - Now the main goal is to prepare our serious scientific work, research, so that it appears at leading conferences such as NeurIPS or ICML-conferences on ML, which take place in different countries of the world. For a career open question, look at how ML has been developing over the past 5 years. It is changing rapidly, now it is difficult to predict what will happen next. And if we talk about ideas and plans in addition to scientific work, then perhaps I would see myself in some kind of my own project, a startup in the field of AI and ML, but this is not accurate.

    - In your opinion, what are the limitations of AI technology?
    - Well, actually speaking of AI as a thing that has some kind of intelligence that processes data, then, in the near future, this is some kind of awareness of the world around us. If we talk about neural networks in computer linguistics, for example, we try to locally model something, such as language, without giving a model an understanding of the context about our world. That is, if we can put this into AI, we can create interactive models, chat bots that will not only know the language models, but will also have an outlook, know the scientific facts. And I would like to see this in the future.

    By the way, the Academy of Artificial Intelligence is currently recruiting students for the new hackathon. The prize money is also solid, and this year’s task is even more interesting - you will need to build an algorithm that predicts the player’s experience based on the statistics of one Dota 2 match. For details, follow this link .

    Also popular now: