Women’s learning: datasentist girls talk about machine learning, career growth and trends

Data Scientists are analytical data experts who have the technical skills to solve complex

tasks. They love mathematics, are almost computer scientists, adore statistics, and most importantly - data and their analysis. In Russia, according to Headhunter, the salary of a specialist in Data Science and machine learning can reach 300 thousand rubles. It is clear that such specialists are very popular and highly paid in the market.

Alexandra Murzina, Machine Learning Engineer, Advanced Technologies Group, Positive Technologies

One of the most promising technician inmachine learning is now reinforced learning. It is on her, by the way, that the DeepMind system is based, which won in StarCraft II. Such an approach to learning, and then to use, really looks more like an AI. Such a system is immersed in an environment that gives it responses to actions. This method is very similar to how we study with you, but sometimes it takes years for us to do this, and here it is possible to significantly speed up the process through modeling and computer power.

For example, such a system will quickly learn "wellto drive a car. Of course, there will remain issues related to exceptional situations and the operation of the system in their conditions (due to the limitation of the tasks it solves). In the autumn of 2017, capsular neural networks caused no less noise: at that time they spoke as much as about a revolution in the world of deep learning. Today they almost forgot about them in public. In practice, still very popular are the busings and neural network architectures, which have already become the standard for solving certain problems. Such, for example, as the detection of objects in images or classification of images.

Theoretically, AI as a technology will come faster inthose areas where a person’s work consists mostly of routine actions, or in areas where decisions need to be made quickly based on a large amount of data. That is, the sensational news about how 600 traders were replaced by two hundred programmers can easily become one of ordinary events in the field of automation. Turning to the cybersecurity industry, such automation is likely, for example, in the area of ​​virus analytics, or the detection of attacks. By the way, our team is working on a technology that allows machine learning to detect attacks on web applications in seconds.

In practice, it is possible to predict the industries in which AIit will be demanded most in full force and in the near future is difficult, since its use is associated with a lot of social and legal nuances. Say, autopilot: it is technologically possible to transfer standard flights to autopilot, but how many passengers entrust their lives to an airplane without a person sitting at the helm? Or, say, medicine ─ there are a lot of developments based on the use of AI in relation to this industry, but from the point of view of the legislative base they cannot be fully utilized and today they remain at the conceptual level.

Yes, there is a lot of noise around the direction: however, many experts prefer to immediately talk about solving specific problems, saving time and money, subject to the use of technology. In reality, these ideas stumble over the personnel question: if a few years ago there were actual talks about the shortage of programmers, then nothing has changed, except that now not just java-programmers, but also complex engineers, are needed now, who can understand and solve the problem, including methods of machine learning.

The feeling of a glut in the marketmachine learning is more deceptive than real. Yes, many people think that they can quickly explore this area after completing a couple of courses, and as a result, the market is oversaturated with specialists with an irrelevant background. However, machine learning is in most cases a tool for effectively solving a specific task (well, only if you don’t do it for its own sake) and in order to choose the right tool you need to have extensive expertise in a particular field (in our case, in information security) .

The history of machine learning today is moreIt resembles the situation with the “golden hammer” antipattern popular in the development environment, in which any problem is tried to be solved with the help of a single (albeit golden) hammer. Machine learning specialists who have completed a couple of courses apply their experience in a hundred cases out of a hundred, without being able to understand when another toolkit is needed — an additional one. Many of these freshly baked specialists are not developers and cannot go beyond the framework of Jupyter Notebook (a popular tool in ML), or lack adequate knowledge in other areas, which does not allow them to successfully use technology in any particular area to solve specific problems.

Alena Arykina, Data Scientist Sberbank PJSC

In machine learning there is a creative part,depending on the data and knowledge of the field, on the intuition of the developer and much more. And there are automatic things, where you need to find better parameters and rewrite a long-known code. The second part, as in any "mechanical" process, people tend to optimize everything, including through machine learning. If earlier mathematicians moved from “manual learning” to automatic one on the basis of such libraries as keras, today examples of such optimizations can serve as libraries for managing the state of datasets, preprocessing pictures and texts, even automatic detection of features of elements. Algorithms can be simple (cut off the endings of words) or complex (build special neural networks - autoencoders that compress data to any size), but a set of such training tools most often determines the quality of the final model, and therefore the skills of a datassentist.

Machine learning will eventually come later in allareas where it will be slowed down by legislation or human distrust: medicine or a car with an autopilot. We already see tremendous achievements in these areas - they are being used in other countries. But I am sure that in order to implement them with us and make them accessible to ordinary people, we will have to win more than one bureaucratic war.

Data Science is really very interesting. Every second my friend IT specialist already at least read about machine learning. Willy-nilly, you start to worry: won't there be too many datasintists? In addition, now they are trying to use machine learning in any IT project and for any tasks, not really imagining why such algorithms are needed there - it is fashionable. HYIP will pass and the number of vacancies will decrease. On the other hand, the question is whether those who really love Data Science will remain in the profession and not chase after fashion.

Tatyana Savelyeva, head of unstructured data group, Yandex.Taxi, author of the tldr_arxiv telegram channel

I don't really like the word “AI,” becauseit is too general and ambitious, and often leads people to overestimate the level of technology. You know, there is such a joke: “How to distinguish ML from AI? ML is done in Python, and AI in PowerPoint. ”

The first trend of Data Science is increasingthe popularity of the subject area: companies are increasingly aware that without processing a large amount of useful information in the future it will be hard. There is a trend in machine learning automation: 10 years ago you had to write all the methods for working yourself, now there are a lot of convenient ready-made libraries.

But with the increasing usability of the methodsActual tools are changing faster and faster - you need to keep your finger on the pulse. There is a trend in the use of neural networks: industrial conferences publish more and more articles related to this type of algorithm.

So machine learning is lastwill come to areas where there is little data or where they are not at all - for example, in this way you can hardly predict the place where the asteroid falls, or the time of the collision of the Moon with the Earth. It seems that machine learning is hard to implement in bureaucratic institutions - government agencies, medical institutions.

In any case, at some point the market willto appear in a large number of applicants for starting vacancies - junior specialists or interns, as the knowledge necessary for finding a job for such a position is becoming more accessible. But the demand for experienced specialists who have already implemented ML projects will grow, since it takes a lot of time and effort to obtain this experience, and the number of machine learning tasks is growing faster than the number of people who managed to get this experience.

Emily Drahl, Data Analysis Analytics, Mechanica AI, Head of Data Mining in Action

In machine learning one of the mostbright trends is the transition from its use as an assistive technology to full automation based on it. This is most clearly manifested in the automation of industrial production, agriculture and agribusiness, as well as the development of the concepts of smart city and smart home.

Now machine learning applicationsquite a lot and this is due to the current level of technology stack development, the level of our understanding of the field and a number of unresolved ethical issues. My personal top application is medicine, psychology and pedagogy. Here, it is not primarily about ancillary services (recommendation systems for diagnosing diseases or interactive systems), but about the full automation of processes through AI and ML.

I think the IT industry today is different becausetechnologies are developing very dynamically and if you stop keeping pace with these changes, then there is a very tangible risk of becoming an unclaimed specialist. This is one of the few areas where university graduates without experience can compete with experienced professionals.

Thanks to the dynamic market, work for those whoKeeps up with trends, will always be. But for those who are not ready to learn all their life, the difficult question remains to be solved: how to remain relevant. Here will help experience, professional outlook and knowledge of related (or not so!) Areas of activity.

The field of education is currently changing.conceptually and, if I may say so, turns around not only to schoolchildren and students, but also to adult specialists with work experience. Having relevant relevant education in the past, a fair amount of time and proper level of perseverance, you can retrain yourself without significant financial investments and have an interview at least on the initial position in the field of data analysis. This is one of the goals that online courses set for themselves.

If we talk about universities, most of themis experiencing a number of difficulties in teaching relevant technical disciplines: technologies change very quickly, you need to attract practitioners, and they are not always ready to work in the format that the university implies. This is how leading IT companies come to help, who create schools, open departments at universities, organize practical courses and internships, as well as train graduates inside the company at the start of their work. In the final account, the task of the university I personally see is not only and not so much to release a ready-made specialist to the market, but that a higher education should give a person a certain cultural, intellectual and emotional level on which his professional life depends to a greater degree rather than knowledge of specific technologies.

Anna Voevodskaya, machine learning expert, Jet Infosystems

It seems to me that now they are applying more and more.reinforcement learning. The decision to learn, interacting with the environment, using rewards, actions and observations. One of the most famous examples of reinforcement learning is AlphaGo. Also, such training methods are used to simulate the movement of a person (the last competitions at NIPS were about RL), machines and other.

Machine learning is magic at its best.sense. Quite a complex mathematics is applied to your data, a deep analysis is done and a very accurate forecast is given just for you. And everyone wants this magic for themselves: it earns money, and it is useful for the image - it’s great.

As for the glut of the market, I'm in thisI do not believe. Good specialists are always hard to find. For example, Java did not appear two years ago, but senior is still hard to find in this area. And a good datasaytnist is generally like a unicorn: he knows and loves math, and kodit, and understands business metrics, and explains everything well. If we have an oversupply of such people at some point in the world, it will be nice. But this is some kind of utopia.