‘When You Have a Lot to Do, You Find Time for Everything’

Egor Churaev specialises in neural networks. In an interview for the HSE Young Scientists project, he talked about his program for determining the emotions and engagement of online conference participants, his trip to Brazil, and his sports hobbies.

How I Decided to Pursue Science

In my third year of undergraduate studies in the Department of Applied Mathematics at Nizhny Novgorod Technical University, I began researching internal waves, mathematical modelling, and numerical methods. During my master's, I joined a robotics team.

In 2016, I considered applying for a PhD programme but did not pursue it. Two years later, I switched jobs, moving from Intel Co. to a small start-up. At that time, machine learning technologies were booming, and I decided it would be beneficial to gain knowledge in this area, especially since it would also help in my work. So, I sought guidance from top experts.

I had heard a lot about Andrey Savchenko. I emailed him, asking if I could work with him during my PhD studies. So, we started with a project for Huawei, which I initiated and ran for several years. Recently, I passed my final PhD exam, and I will have to defend my dissertation soon.

What I Study

The topic of my PhD dissertation is ‘Personalised Models for Recognising Psycho-Emotional States and Engagement Based on Facial Video.’ I also developed software that tracks changes in psycho-emotional states and engagement in real-time. Our goal is to ensure that the models operate efficiently across a wide range of devices while providing meaningful feedback.

For example, online instructors often struggle to assess student engagement—how attentive they are during a lecture or whether they find it interesting. The level of student engagement directly impacts how well the material is absorbed. Real-time feedback significantly improves the quality of online lectures by enabling the material to be adapted to the audience.

How I Worked on My Thesis

The models we use were developed by my academic supervisor. They are designed to identify emotions in images. Andrey Savchenko suggested I train the model on his dataset to replicate results from his previous work.

I later found a public dataset and achieved high accuracy in emotion detection—almost 98%. Initially, I thought this was correct, but my supervisor and I realised there was an issue with data splitting during training.

The same person could appear in both the training and testing datasets but with different emotions. Because the model already ‘knew’ how I express anger, for instance, it could more easily identify how I express joy.

This issue was present in many studies using the same dataset. We addressed it by applying a personalised approach.

However, not all initial ideas were implemented. For instance, we wanted to use a multimodal approach, incorporating audio along with video. However, when Sberbank approached us with a request to determine audience engagement, we focused on that and did not pursue audio-based emotion recognition further.

About the EngagementEstimator Program

My program, which can detect engagement and emotions, is called EngagementEstimator, and it is available on GitHub. It serves as a demonstration of algorithms that can be adapted for various purposes. For example, the HSE AI Research Centre is currently developing its own framework to unify the research of various organisations within the centre. This framework is designed to showcase all inventions related to the centre. The algorithm implemented in our program has been integrated into this framework and is now operational there.

What Makes Me Proud

This summer, with financial support from HSE, I was thrilled to attend a conference in Brazil, where I presented our research. Our study was published in a top-tier CORE A conference.

However, I find it difficult to evaluate my own achievements. To the average person, it might seem amazing that a computer can read emotions. But once you understand how it works, it does not seem that complex. We trained the model, and it began producing decent results. We identified a problem and proposed a personalised algorithm. The key points I am presenting for my defence are precisely these personalised algorithms: one for emotions and another for engagement. Additionally, we proposed an algorithm for real-time model fine-tuning, whether for engagement or emotions. During real-time operation, we collect video data and ask the user to evaluate the model's predictions. The user can correct these predictions, and we record this feedback to fine-tune the model. This process ensures higher prediction accuracy in subsequent use. These three algorithms—personalised emotion recognition, personalised engagement detection, and real-time model fine-tuning—represent significant scientific contributions.

How I Became a Developer

I started programming before computer science lessons began in school, thanks to my father. He told my twin brother Ilya and me that if we wanted to play computer games, we first had to spend half an hour doing something useful. Programming drew me in: I developed my school’s website and attended Intel seminars during my first year at university. As soon as I could, I started working there.

How I Deal with Burnout

I do not have any burnout per se, but I sometimes feel imposter syndrome. For example, my supervisor defended his doctoral dissertation in his early 30s. I am amazed by his efficiency and feel far behind. However, attending academic conferences helps. When I listen to other presentations, I realise my scientific work is on par with other researchers.

My Hobbies

I enjoy sports, programming, travel, photography, and books. When you have a lot to do, you find time for everything. Before the conference in Recife, Brazil, I arrived a week early to explore Rio de Janeiro, visit Iguazu Falls, and even travel to Argentina.

I also rock climb and play volleyball. You cannot sit at a computer all the time; you need to switch gears.

Advice to Aspiring Scientists

Do not be afraid to experiment and trust your academic advisor—they will not give you a bad piece of advice!