The Creators Of Artificial Intelligence - 4 Stories Of AI And ML Specialists
Four experienced specialists on how they dealt with artificial intelligence, what difficulties they encountered and what tasks they solve.
“At first it was always scary, but I never regretted”
The co-founder and technical director at the startup, is responsible for technological development, architecture of solutions and evaluating their quality, application of AI and other technical issues, and is involved in managing, developing and hiring employees.
 |
| Credit @AIMagnus |
The beginning of the journey . It is difficult to say what exactly led me to the profession: from childhood I was interested in programming. At the same time, I was interested in psychology, biology, mathematics, radio electronics, read the magazines “Young Technician” and “Young Naturalist”.
For a long time, everything related to artificial intelligence was for me a hobby rather than a profession. At some point, I realized that these topics also make up a significant part of my work tasks.
First difficulties . In my case, the transition was gradual: my professional career began with various IT projects, and the first “commercial” project was the search engine for the Moscow Abstracts Collection written in Perl. It was not always possible to understand new topics the first time, I had to re-read several different books in order to understand something, and also experiment a lot.
I repeatedly got involved in completely new projects, where at the start I completely lacked the necessary knowledge - I had to master along the way. At first it was always scary, but I have never regretted it.
To understand how everything works, almost always I started programming from the lowest level. So it was with the Bresenham line drawing algorithm, Fong or Gouro shading — when I studied computer graphics, and with the creation of a simple neural network, the implementation of the support vector method or the genetic algorithm — when I dived deeper into AI. Then I went over myself for a long time: I did not want to use ready-made libraries and tried to write my own from scratch.
Professional tasks . Artificial intelligence has become a fairly versatile technology. Over the past few years, with the help of ML or Software Engineering, I and my colleagues have done very different tasks:
Image Recognition: traffic signs from a smartphone’s camcorder or goods on a store shelf;
structuring the news flow:
clustering news on common topics, annotating the resulting clusters and highlighting important facts, ranking the flow by importance and the like;
forecasting in education: which of the students will drop an online course in the near future;
Realtime call center analytics: determining the topic of a telephone conversation and people's emotions;
analysis of genomic data: to determine the structure of chromatin;
work with texts: finding sentences matching each other between parallel texts in two different languages;
Now I identify the weaknesses and strengths of models and services based on artificial intelligence. This helps to choose which ones are suitable for a specific business task.
Future plans . What do I plan to do next? I will apply my acquired skills in the fields of medicine and biology, study the “psychology” of natural and artificial complex systems, try to create an AI scientist, or at least an assistant, to increase my efficiency. I also plan to learn several new programming languages: Rust, Swift, Kotlin, Julia or Elixir. I’ll also try to make more hardware projects with artificial intelligence based on Jetson Nano, Google Edge TPU or with FPGA.
Dmitry Korobchenko, Deep Learning R&D Engineer and Manager, NVIDIA
The head of the R&D group, is engaged in image processing: the use of neural networks for image processing, computer graphics, animation and physical simulation.
The beginning of the journey . While studying at the university, I was fond of computer vision and therefore decided to join the Laboratory of Computer Graphics and Multimedia at the Faculty of Computational Mathematics and Cybernetics of Moscow State University.
Working at Samsung after university, I returned to computer vision: one of my first projects was the analysis of medical images using convolutional neural networks. And when in 2012 neural networks actively spread to other areas, the range of my projects expanded significantly.
Professional tasks . As a Deep Learning R&D Engineer, I am engaged in both research and development: from creating new algorithms and conducting various experiments to the implementation of final products with subsequent optimization. In addition, for the last few years I have been giving master classes and being a teacher in machine learning and neural network courses at various additional education schools.
Now most of my tasks are related to complex data types - images, sounds, polygonal models, tensor data and so on. Including I continue to be engaged in computer vision: image classification, object detection, semantic segmentation; create neural network frameworks.
Future plans . My immediate plans are to strengthen the R&D group specializing in neural networks at the NVIDIA Moscow office, as well as continue to develop in the educational field in the field of artificial intelligence: to make content for lectures, courses and a personal YouTube channel.
- Learn to put forward ideas and hypotheses and create a plan for solving the problem
- Learn how to select algorithms and metrics for a task for different models
- Gain applied experience in creating working neural networks instead of a “top-down survey”
“At the dawn of the 2000s, the term Data Scientist did not exist, but in fact I did just that.”
Anna Kostikova, Director Data Science and Bioinformatics at Novartis
He leads a team whose tasks include creating personalized medicine in the development of new drugs. The essence of the group’s work is that drugs are developed and selected based on the analysis of digital information about DNA, proteins and clinical data of patients. To do this, Anna and her team use machine learning, bioinformatics and statistics.
The beginning of the way. At the dawn of the 2000s, the term Data Scientist did not exist, but in fact I did just that. For example, at the university I found a side job where I had to collect data for a database, come up with a structure and make it possible to work with the database. All this I did in MS Access on a computer with 512 MB of RAM and 1 GB of hard disk space🙂
In the third year, I got an internship at a non-profit company, where they were engaged in the analysis of space images. It was then that I first tried to apply neural networks, classification without training and with it, fuzzy logic and so on. Then computers with 4 gigabytes of RAM were comparable to a miracle, and we did not turn them off for the weekend - so that they “counted” while we rested.
First difficulties. For the first time, working with the official “title” Data Scientist came to me in 2014. Then I got a job at Booking.com and found out what it feels like to work in this area on an industrial scale: with data samples of billions of rows.
The first two years are the most difficult in any field: all the terminology is new for you, it is not clear what is important and what is not.
Learning new things is always a sigmoid function: you must overcome the first plateau when it seems that you will never understand. For example, in a graduate school in Switzerland, I needed to learn how to analyze genomic data and write a Perl script for large-scale analysis. At that moment I did not know any of this, but somehow I got out. The main thing is not to be afraid and try.
Professional tasks. In my practice, there were a lot of different tasks: from analyzing satellite images for WWF to optimizing the brewing process in Heineken, from predicting user behavior on the Internet for Booking.com to predicting the functioning of drugs in Novartis.
I currently work as Director of Data Analysis and Bioinformatics at Novartis. I also have my own cancer diagnostics company. I would really like to make the most of Data Science and machine learning for healthcare and medicine - from drug development to diagnosis. I believe that for the next 20 to 50 years, the lion's share of the efforts of analysts around the world will be aimed at solving biomedical problems, changing the quality of life of mankind, and not just optimizing on the Internet and in production.
“Then I was sure that I already knew everything, although I still knew nothing at all”
Nikita Semenov, NLP team lead, MTS artificial intelligence center
He directs the NLP teams and deals with everything related to the processing and understanding of a natural language.
The beginning of the way. Even in the first years of the institute, I began to optionally study machine learning: I studied the specialty "Computer Security", but gradually I realized that I would not want to connect my life with it. My research supervisor graduated from the Milan Polytechnic with a Computer Science program, and with him we began to develop an optional machine learning course. A similar term did not exist then, and all over the world they talked only about the elements of static learning, which we studied. Unfortunately, in Russia there are still no such programs on Computer Science.
It was extremely difficult to find a job after high school precisely by the profile of machine learning - the sphere was just emerging. So I went to a small startup that was engaged in automating bids on contextual advertising sites like Google AdWords. My first task was to develop a mechanism based on statistics and the predictive ability of the goal in such a way that we always occupied not the first bid, but the second or third - these lines also show in the output top, but they are much cheaper. Then I was sure that I already knew everything, although I still did not know anything.
First difficulties. Soft skills were a big difficulty for me: I had to explain what I was doing, what it all meant, how and what to interpret, and what the effect would be, to people who did not understand anything in my field. At that time, peer learning processes were not yet mainstream, so it was very difficult to interact with the team. I constantly practiced: I tried to convey my thoughts and explain to the team even the simplest metrics. I think if I was just starting my career now, I would not be able to pump out so much in communication - almost no one asks such questions.
There were no problems with hard skills: then my tasks were based on statistical training and mathematics, which I was well versed in. Despite this, I still read books: in Data Sience, you need to constantly develop in order to understand the tools and trends. In general, all my work experience is one big challenge. Each place required new tools and knowledge, therefore it was always necessary to develop independently.
After working in a startup, there was a company where I became the first Data Scientist and R&D: I helped set up the first analytics tools, was engaged in computer vision and the construction of predictive models based on data from space satellites.
Professional tasks. In MTS, I came to the position of Senior computer vision engineer, and then grew to the team lead of two teams. It is especially important for me to pump soft skills, because team leader is a playing coach. If we talk about the tasks, then here I do everything related to the processing and understanding of the natural language. Now this is a kind of trend, which sets new trends aimed at simplifying human life in the future.
Over time, I realized that the subject area does not affect your area of knowledge so much. In my case, the subject area always affects how to process and apply data to any solution. And the approaches always remain the same. And when in the subject area, specialists come up with an innovative solution, for example, in computer knowledge, over time it flows into other areas. In this regard, the line between the regions is gradually erased, and approaches and bases are becoming similar.
The main problem of our sphere is that it develops very unevenly. Let me give you an example: in Data Science, for a long time, everything can be calm, and then someone sharply comes up with a solution, and after a short time, these breakthrough things become the standard for everyone. In terms of work, this is both good and bad at the same time: on the one hand, you constantly pump skills and “run” 10 times faster than the others, on the other hand, your work profile is constantly changing.
Plans for the future. So far I have no understanding in which areas I want to develop further. I want to dive even more into what I am doing now.
Most experienced AI professionals came into the profession in one of two ways:
- moved from another field of activity, profession;
- worked on other specialties in IT, and at some point plunged into tasks related to artificial intelligence and data analysis.
In any case, even experienced specialists are constantly learning new things, studying useful resources and articles, and taking advanced training courses.
Despite the fact that Russian universities have not yet implemented the appropriate educational programs, it is now easier to become a Data Scientist than during the heroes of our article.
If you are interested in the field of artificial intelligence, machine learning, and analytics, we invite you to study the programs of our Deep Learning and Machine Learning courses .
No comments: