NYU Center for Data Science, Meet the Researcher: Elena Sizikova

Elena Sizikova
Photo Courtesy of Elena Sizikova

Meet Elena Sizikova. Before joining CDS as a Moore-Sloan Data Science Fellow, she acquired her PhD from the Graphics/Vision Lab in Princeton’s Computer Science department. Her research interests are in applications of computer vision to medical imaging and perception, 3D reconstruction and visual understanding tasks. She is also passionate about encouraging diversity and minority participation in computer science and mathematical research as well as developing online learning tools.

We caught up with Elena to discuss her research projects and background, what initially sparked her interest in data science, and her overall experience at CDS.

This interview has been lightly edited for clarity.

I see you received your PhD in the Graphics/Vision Lab in Princeton CS department. What brought you to CDS?

I graduated from Princeton in 2019. I was there for six years and mostly worked on research in computer vision. From there, I wanted to go to a university where I’d be able to study applications of computer vision in healthcare and medical sciences. I’m very interested in medical imaging as well as AI applications in healthcare. I look at ways how neural networks and computer vision techniques can be used in these domains. In addition to more traditional computer vision research, I’ve also worked at Siemens Healthcare where I was able to see how these techniques are used for practical purposes. From a research perspective, I joined CDS because it was a place where I can work on both theoretical research as well as applications by collaborating with technology. From a personal perspective, I always wanted to live in a big city, and I absolutely loved New York so it’s a great fit for me.

How has your experience at CDS been?

There’s a ton of different things going on, both research and non-research related. But especially in terms of research, I’ve been able to start collaborations with several professors and students. I’ve encountered a lot of really smart students who want to try out new ideas and are keen to learn new things, and it has been great to work with them. Directly advising students was a really eye-opening experience for me, quite a transition from being a student myself. It’s been great. I love seeing both sides — I was mentored before and now I get to do the mentoring.

When did you first become interested in data science?

It started in my high school and undergrad. In high school, I participated in a lot of math competitions that basically brought me to study math and science. I’ve always felt like I didn’t want to only work in a theoretical field, but study how theory and applied math or computer science can be interfaced. I was interested in the applications of math, computer science to biology and language. In my opinion, both theoretical and applied fields can complement each other in their goals.

Tell me a little bit about the research you’re currently working on?

One of the goals of my research is to study the differences between how humans and neural networks learn. For example, if we wanted to build a computational model of reading, we would want to see how font size, image noise, and other factors affect speed and accuracy. With Professor Denis Pelli and his lab, we are building a computational model that would explain both slow and fast reading from visual inputs, and try to model dyslexic reading with a neural network. With my independent study students Sahar Siddiqui and Diksha Meghwal, we are analyzing how neural networks make mistakes, and how this information can be used to improve them. In an ongoing collaboration with Princeton University and Google researchers, we are studying how can we design the right templates to manipulate shapes in a human-meaningful way.

I am also interested in how mathematical tools can be used to interpret and improve computational models for tasks such as object classification and localization. I am working on developing an interpretable model for COVID-19 patient X-rays, with NYU School of Medicine as well as collaborators from UCLA and Occidental College. We’re trying to build an interpretable model of X-rays. We are analyzing where a patient would need a ventilator, intubation or ICU treatment, and detecting how each region of the X-ray points to that prediction. I believe we’ve seen about 5,000 X-rays so far. I am also trying to create better models for identifying objects in images, without human input (known as the unsupervised object discovery problem). For example, if you have a bunch of images, and you’d like to locate the sample object in all of them. With Professor Jean Ponce, PhD student Huy V. Vo, and Valeo.ai’s director Dr. Patrick Perez, we are improving the mathematical algorithm used for solving this problem so that we can scale to large datasets. Finally, I am working with a PhD student in statistics at Southern Methodist University, Yuzhou Chen, on applications of topological data analysis (TDA) to geometric deep learning.

What role do you see your work potentially playing in the future of Data Science?

In the one of the projects I mentioned, we’re building a technique for interpreting machine learning and machine learning models that are human-like or more human-like to prevent them from making stupid mistakes. And by figuring that out, I hope that we can actually make machine learning that is used for medical imaging and healthcare more practical. With existing models, we want to understand what the errors are and why they so people will have more trust in them.

I also hope to bring more awareness to the different parts of data science and how it can be used as a “glue” between theoretical and applied problems, as I believe it is a crossroads of finding similarities in different fields and opening new paths. As a female researcher in data science, I also hope to be a positive role model, so that other women and members of underrepresented groups in computational sciences that are considering research careers in data science or computer science, can be empowered to do so.

Any final thoughts you would like to share about yourself, CDS, or data science in general?

I really enjoy my time at CDS. For example, when it comes to organizing student projects, I’ve been mind-blown by how we’ve been able to get speakers from all kinds of different industries and backgrounds. New York is also a great place to be, it is “a world within a world”.

By Ashley C. McDonald

Official account of the Center for Data Science at NYU, home of the Undergraduate, Master’s, and Ph.D. programs in Data Science.