Yandex Research Machine Learning Residency

We announce the start of the Machine Learning Residency program — a position for early-career professionals designed to provide experience in top-tier machine learning research. Our ML Residency program is well suited for young researchers on the path to a full-time role in the industry or academia. The program is organized by Yandex Research in collaboration with MIPT and HSE University.

Residents are offered an opportunity to engage in applied and theoretical machine learning research, gain experience working on high-impact projects, publish at top-tier academic conferences (NeurIPS, ICML, ICLR, CVPR), contribute to high-tech production services and collaborate extensively both within Yandex and with top universities and industrial labs in the world.

Residents are paired with senior Yandex researchers who act as mentors. Together, they choose a research problem from areas such as:

Computer vision;
Speech and dialog systems;
Natural language processing;
Autonomous driving;
Probabilistic ML;
Reinforcement learning;
Graph ML and others.

Their work may result in publication at conferences or journals, implementation at Yandex or release as an open-source project.

We are looking for candidates with a strong technical background, a genuine enthusiasm for ML and a desire to have a real-world impact. Residents will receive a competitive salary. The ML Residency program is an evolution of the Yandex Academic Supervision Program that was designed for master's and PhD students. Now we also welcome experienced researchers from a background outside pure ML, such as mathematics, physics, or computer science.

Here are some recent research projects done as a part of the Academic Supervision Program:

Ivan Provilkov and Andrey Malinin, Multi-Sentence Resampling: A Simple Approach to Alleviate Dataset Length Bias and Beam-Search Degradation, EMNLP 2021.
Michael Diskin et al., Distributed Deep Learning In Open Collaborations, NeurIPS 2021.
Dmitry Baranchuk, Andrey Voynov, Ivan Rubachev, Valentin Khrulkov, Artem Babenko, Label-Efficient Semantic Segmentation with Diffusion Models, ICLR 2022.

Please visit this page to learn more about the program and to apply to ML Residency.

Personal stories The link has been copied to clipboard

We asked several researchers who participated in the Academic Supervision Program to share their experiences.

Dmitry Baranchuk

I joined Yandex Research at the end of my third year at Lomonosov Moscow State University's Faculty of Computational Mathematics and Cybernetics. Since there was no ML Residency program back then, I applied for an internship.

I wanted to do research at the level of the strongest laboratories in the world, only I was an inexperienced student who didn't know how to search for a research topic or quickly decide which of my ideas was viable. My mentor was wonderful, and we worked closely together. When we first started, he set a direction for our work and came up with ideas I tried to quickly check. I gradually began to generate interesting thoughts as well, making my own small contribution. Under my mentor's guidance, I followed the logical path from intern to researcher.

Yandex Research represents a wide range of possibilities. As I tested out a variety of areas in machine learning, I was able to participate in successful projects that resulted in publications. I remember my first ICML conference in Los Angeles. It was fantastic! I felt like a superstar. Just imagine: you're a regular student, a kid who comes to Los Angeles and speaks to a huge audience of some of the smartest people on the planet. The experience was unforgettable, and I was stunned by the scale of the scientific and academic world... Without Yandex, it would have taken me far longer to achieve what I have.

I'm currently researching advanced image generation models, leveraging them to solve practical problems in computer vision. For example, my last project was to dig into one of the models to see how well it understands the structure and semantics of the images it generates. The understanding turned out to be quite good, and that knowledge can be used to highlight semantically similar objects in real images. And it is noteworthy that the model needs just a few annotated pictures to do it.

Who is Yandex Research right for? People who are constantly coming up with new ideas and trying to understand everything at the deepest possible level as well as people who don't have much research experience but really want to be a part of the scientific community and share their ideas with a wider audience.

Michael Diskin

I'm studying at HSE University. Before joining the ML Residency program, I worked for a large international company as an ML researcher. But it was a boring position. I like Yandex for the intelligent people and the interesting work we do here.

I came to Yandex Research with a few goals: work on my research skills, delve deeply into modern ML, and learn how to work more effectively. And I achieved every one of them. Not only am I much better versed in my field, but I've also moved into related ones by constantly working with colleagues and going to seminars where they share interesting papers. That helps me keep in touch with what's going on around me.

One thing we're doing in the research department is developing a library that helps people train huge ML models together. By "together," I mean that if one person doesn't have enough computing resources to train their model, they can team up with other people to solve the problem.

The most valuable thing for me is my relationship with my supervisor. Firstly, he's flexible enough to let me come work whenever I want. Secondly, this is the only time I've ever been able to trust my supervisor. Many people don't realize how important that is. No matter the moment, I can ask for his opinion and get an honest answer about myself, a project, life, or anything else. And I can be sure he's not going to think I'm dumb, be condescending, or get rid of me with some platitudes.

Who is Yandex Research right for? I'm working in a field called distributed learning. Not just your typical machine learning, it's a fascinating combination of engineering, coding, ML, and ML experiments. I think that might interest a lot of people. Other Yandex Research departments are occupied with all things modern ML: computer vision, program generation, distributed learning, different disciplines in mathematics, NLP, and more.

Yandex Research generates half of Russia's scientific publications on machine learning. If you're looking for real experience in research that's advancing science instead of just gunning for a slick PR move, this is the place for you.

Ivan Rubachev

This is my second year working in science at Yandex Research. I am also completing my master's degree at HSE University, which is where I learned about this program.

It all started by accident. Science has been an idea spinning around in my head since my second year. And as my fourth year was coming to an end, it turned out that I can enroll in the master's program, which was my plan, and join Yandex Research's scientific internship program (now ML Residency). I decided that the moment had come. After passing the standard selection process, I entered the master's program and joined Yandex Research at the same time.

But it wasn't easy. In the beginning, my job was tough and introverted. I was in quarantine and couldn't see my colleagues. Back then, I didn't know how much isolation can impact your work. When you're sitting at home, it feels like your supervisor is superhuman, while you're just some intern off working your way through your assignments. That's not a winning strategy. When I finally got to the office, everything turned right side up, and I enjoyed the great collaboration we have.

I feel like I'm where I'm supposed to be. Thanks to Yandex Research, I've developed mechanical field skills: how to set up experiments correctly, how to accelerate processes, how to write papers, and how to code better. If it weren't for the program, our paper on NeurIPS would never have happened, and I wouldn't have started giving seminars on deep learning at HSE University.

I'm now wrapping up my master's degree, so we can say that the program as it was meant to be is coming to an end. I'm happy with what I've achieved. Back when I first joined, I was really an intern. I remember asking myself, "Are there any simple metrics in science? I have to write papers. Writing a paper would be great!" Now our team has two of them, and I'm happy.

Who is Yandex Research right for? If you're like me and the idea of science is buzzing around in your head, if you feel like you have that inner light burning, you should step out of your comfort zone.