Skip to main content
Digital Environment Research Institute (DERI)

Ms Tatiana Gaintseva

Tatiana

Computer Science PHD Student

Email: t.gaintseva@qmul.ac.uk

Profile

Project: Computer vision, diffusion models, explainability and knowledge extraction
Supervision:  Prof Greg Slabaugh and Dr Ziquan Liu

Why do a PhD?
Because I want to be a researcher, this is the only job that interests me.  And the first step is to complete a PhD :)

What are you researching and what led you work in this field of research?
I am interested in exploring intermediate representation of models and interpreting their behaviour. I work in this field for several reasons,  some years ago at the beginning of my career I was searching for a topic in AI research that would interest me and started working as a researcher in computer vision, building better models for representation learning. I tried to create techniques which would make models show better results, and while working at this I became interested in better understanding inner processes of the models, as I believed that with such understanding I would be able to create more targeted and justified ideas on how to make model work better.  I also personally love to delve into the details and mechanisms of everything, discovering new connections and finding new patterns. That's how I understand what I love to do in my research.

Who has influenced you along the way ?
I admire people who have a true desire for their work. Who do not just follow latest trends and try blindly stick to it, but who are certain in their intentions on work. Among great people I would say I admire Geoffrey Hinton  and Yann LeCun, as I see how they express their ideas regarding AI and work on them. These ideas may not be the most commonly recognized ones and may not be related to the latest trends such as GenAI, but they follow their path and their excitement is clear.

What impact do you want your research to have?
I want my research to provide useful insights about how AI models work, so that new, better models and better techniques for solving different tasks could be build on top of that knowledge. I also aim to provide interpretable methods of solving different tasks. 

What is the most intriguing puzzle that your research has revealed so far?  
That intermediate representations spaces of different models have different structures with intriguing properties. And it is possible to control the behaviours of models in quite precise ways by easily manipulating these intermediate representations.

What online resources do you recommend? 

  • Lil'Log blog:  This is a blog with great articles on different AI-related topics. They are written with clarity and depth, with interesting and useful insights.
  • Jay Alammar blog: Here you can find articles with explanations of how different things in AI work. Explanations are visual and of a great clarity, many other articles on the internet refer to Jay's blog.

Have you any key tips on keeping  motivated throughout a PhD?
I am motivated by my inner desire to do research and solve mysteries behind work of AI models. I do believe that true desire for work is essential component for researchers and for PhD students, this is what gives one a true motivation. However, even with such desire there will be moments when something doesn't work, papers are rejected, and this can be disappointing. That's why I believe that it is important to have some other activities outside research, which will boost your confidence and mood. For me it's teaching: I teach Deep Learning, I love it  and I'm good at that. This helps me level up my mood and accumulate strength to continue my research.

Do you have time outside your Academic work for any special interests?
Yes, one of my biggest interests is teaching. I teach Deep Learning, and I love it. I love thinking about how to better explain things, creating methodology for courses, seeing how my classes help people on their paths. Beside that, I love reading books, walking on the streets of cities, playing volleyball and tennis, meeting new people, attending educational events.

Connect with Tatiana Gaintseva 
Google Scholar
Github
LinkedIn

News:
European Conference on Computer Vision (ECCV) success.  We are pleased to announce the following paper has been accepted for publication at the conference, to be held in Milan from 29th Sept through to 4th October.    Congratulations to Tatiana Gaintseva and co-authors Martin Benning and Greg Slabaugh for the paper, "RAVE: Residual Vector Embedding for CLIP-Guided Backlit Image Enhancement," which shows how to use a residual vector in Contrastive Language-Image Pretraining (CLIP) embedding space for image enhancement.

Education Background information:

  • Bachelor’s degree iApplied Mathematics from Moscow Institute of Physics and Technology (MIPT)
  • Master’s degrees in Data Analysis from Moscow Institute of Physics and Technology and Yandex School of Data Analysis (YSDA).
  • My work is supported by DeepMind studentship.

Research

Back to top