News

Get to know Sarthak, Olga and Vésteinn

Featured image

Date

Meet three of the PhD Students working in the Pioneer Centre for Artificial Intelligence.

PhD Student Sarthak Yadav

What is your position and what will you be doing in the Pioneer Centre for AI?

I’m a PhD Fellow in the Signals and Decoding Collaboratory, and I’m at AAU Electronic Systems.

What is your main research interest?

My main research interest is at the intersection of deep learning, signal processing and interpretability. More specifically, I’m working on self-supervised learning of general-purpose audio representations and developing a framework for inspecting such representations.

What are you working on at the moment, and why is it important?

At this very moment I’m working on a self-supervised learning paradigm known as Masked Autoencoders from the perspective of audio.

What are some of the most common misconceptions about AI that you have faced?

I can tell you all about misconceptions I’ve noticed, especially from my industry experience:

1. Comparing Machine Learning and AI to regular software in terms of determinism. AI is stochastic, and expecting complete determinism for test cases like one would expect for regular software is probably the biggest misconception among stakeholders: both company management and end-users.

2. Not understanding that the technology needed to solve a business problem is not mature enough.

3. Too much emphasis on methods and models, and not enough on data, especially in Deep Learning. Believing that a shiny new Deep Neural Network architecture is somehow the answer to all your problems.

Which person in the world would you above all like to meet – dead or alive – and why?

Difficult to choose, so I’d take whoever is available out of the following, in no particular order: Marcus Aurelius, Enrico Fermi, Charles Bukowski, Marie Curie.

What is your favorite thing about living in Denmark?

There are a lot of things I like about living in Denmark: Bureaucracy just works, workplace laws are quite transparent, great work life balance. But, most of all, dressing in all black/grey, all day every day. India’s quite a bit more colourful, and it can be a chore on days I’m not feeling it. It can also be a problem though when I do feel like dressing colourful; it’s a balancing act.
 
 

PhD Student Olga Iarygina

What is your position and what will you be doing in the Pioneer Centre for AI?

Currently, I am a first-year PhD student. In the Pioneer Centre, I am a part of the Extended Reality Collaboratory. With Aske Mottelson and Kasper Hornbæk, I am working on a project that investigates the replication crisis in the field of Human-Computer Interaction in general and Extended Reality in particular.

What is your main research interest?

At the moment, in research, I’m most interested in what new opportunities Extended Reality opens up for conducting psychological experiments.

What are you working on at the moment, and why is it important?

Right now, I am working on two projects. The first project is a large-scale replication study of the fundamental paper about embodiment in Virtual Reality research. This project is important because although XR is well-suited for large-scale user studies and cross-lab replications, in most of the studies, the sample sizes are rather small, and the findings are thus questionable. With this project, we want to introduce such initiatives to XR, as they have improved scientific practice in other areas. My second project is about confounds in experiments that can undermine the validity of the results.

What are some of the most common misconceptions about AI that you have faced?

In my experience, people often have very extreme opinions when they talk about AI. They either think it is perfect and heavily rely on it, or they are subjective to algorithmic aversion, and since they do not understand exactly how artificial intelligence works, they do not trust its decisions. Also, people tend to develop folk theories about AI that subsequently affect their trust and reliance on AI.

Which person in the world would you above all like to meet – dead or alive – and why?

It is not a very scientific answer, but I would like to meet Picasso. I admire the philosophy and the attitude that he put into his paintings and into his life.

What is your favorite thing about living in Copenhagen?

I am in love with cycling! When I moved here, I did not even expect how much I would like it. And also, Danish sourdough bread is very nice.

 

PhD Student Vésteinn Snæbjarnarson

What is your position, and what will you be doing in the Pioneer Centre for AI?
I am a PhD student.

What is your main research interest?
I have a broad interest in natural language processing and computer vision. On the language side, my focus has been on low-resources settings in the context of the Scandinavian languages. On the computer vision side, I have been investigating the capabilities of generative models in rendering fine-grained categories such as animal species. More recently, I have been thinking about properties of shared embedding spaces in models that incorporate both text and images.

What are you working on at the moment, and why is it important?
I am currently working on automating generations of detailed image descriptions. These captions can be valuable on their own, but they can also serve as intermediate steps in further language based processing, for instance for classification when the supporting evidence is only in text form.

What are some of the most common misconceptions about AI that you have faced?
There’s a lot of hype, but also a lot of pessimism in some circles about the large generative models that are in the spotlight. I certainly don’t think the models are getting close to something we would equate with consciousness, as some have claimed, but I don’t think the behaviour we are seeing should simply be explained as machines that do low-level pattern matching. We are far from getting a good grasp of what the trained algorithms are doing, so a healthy dose of scientific caution is apt. When interacting with people who are not working in the field, I think the most common misconception is that AI based methods are continuously learning after they have been put into production use, this is very rarely the case.

Which person in the world would you above all like to meet – dead or alive – and why?
Tough one! The first that come to mind are Bertrand Russell and Johann von Neumann. Both were at the forefront of multiple fields, yet concerned themselves not only with theory or applications, but also complex social questions.

What is your favourite thing about living in Copenhagen/Denmark?
It’s a very liveable city with tremendous commuting infrastructure, it has plenty of attractions, good food and green areas.