Sebastijan Dumancic: “I like to think about problems and formulating scientific questions “

🕒 27 min

Thank you Sebastijan for joining our monthly chat with our alumni. We are happy to have you here as our first postdoc. Sebastijan is currently in his second year of postdoc in computer science, working in the exciting field of AI.

Let’s start with the root cause of your postdoc – love for science. Or love for computers?

I cannot really tell you what was first, love for science or computers. As a kid, I was interested in so many things and I especially loved to read encyclopedias. Believe it or not, I would go to the library to borrow them, but since I didn’t want to be stressed out by the deadlines, I would usually scan the entire encyclopedia to read it in my own time.

Honestly, I don’t think you are the only one. I still remember my own green encyclopedia with dinosaurs. But it seems like you already knew something about computers by then?

Indeed, as I grew up in a family of two informaticians who had a small company focused on computer science education, giving people different workshops on how to use computers or various software. So I knew my way around computers and I already took some elective informatics courses during elementary school, where we did some very basic programming. This was additionally deepened during my high school days.

Education (past – present – future)

Tell us something about your time in high school.

One of my main reasons to go to science-oriented gymnasium (or grammar school) is to get more math and more informatics, but I was mainly interested in everything. I think this was, to some extent, a problem that I had from early on – too many interests, too many things that I wanted to try out. So not only was I going to a regular high school, I was also attending a music school where I played the French horn. I had broad interests and enjoyed pursuing many of them, although it was often funny how I would leave home pretty early with a school bag, a case with my French horn and a sport bag and return home only late in the evening. But I don’t regret it – it helped me develop useful skills for later and also allowed me to easily connect to people from different backgrounds.

• High school: 2005 – 2009, Gymnasium Pozega, Croatia
• High school (music): 2005 – 2009, Music school Pozega, Pozega, Croatia
• BSc: 2009 – 2012, Computer Science, Faculty of Electrical Engineering, University of Zagreb, Croatia
• MSc: 2012 – 2014, Computer Science, Faculty of Electrical Engineering, University of Zagreb, Croatia
• PhD: 2014 – 2018, Computer Science, KU Leuven, Leuven, Belgium
• PostDoc: 2018 – current, Declarative Languages and Artificial Intelligence Group, KU Leuven, Leuven, Belgium
  • Publication Information

So, I assume that picking your field of study was an easy choice. I mean, you were already very into computer science by then. Was that the case?

I wish! I cannot tell you how nice it would have been to have had my mind set on that, but that was definitely not the case. There were days I even wanted to go and study architecture! But as my final two choices I was deciding between studying mathematics at the Faculty of Science and computer science at the Faculty of Electrical Engineering (FER). I really loved mathematics, I loved the abstraction and logic behind it. In the end, I talked to a few people and realized that if I studied mathematics, it would be a lot of theory and proving theorems, while I wanted something more practical.

Tell us more about your studies as an engineer, how was this organized?

Before I delve into details, I want to say that I think that computer science studies have a kind of perception problem. That being that the only contact with computer science a typical high school student has is via programming. This creates wrong expectations, as programming is just a tool; computer science is much more than that. For instance, many students that decide to study computer science do so for the sake of one day making their own video games. So they think it’s all easy, no need to do much but programming. But reality is totally different.

There is a lot of math that one needs to learn, engineering and physics. I mean, you don’t need to be the best at it, but you need to understand logic and mathematical formalism. To create a complex video game, you simply need these things! And because the first two years of computer science studies involved more general subjects with some core math-focused subjects, it is easy to get lost and lose interest. So please, if you’re considering studying at FER or doing any computer science course, do yourself a favor and get ready for some math!

That’s very good advice. But now, out of curiosity, can you exemplify what kind of math one would need for making a video game?

Video games are all about graphics, right? So, the simplest thing you can imagine is the first scene of a video game. Let’s imagine it’s a room in a building with one tiny window, no lights and some furniture. And as simple as it is to draw these objects in 2D, to give the player a third dimension, you need a source of light. In our example, the light is coming from the window, which is where the math starts. How does the light spread around the room? Which parts are lighter, which are dimmer? How far does the light go? How is the shadow cast on the floor after hitting an object? All of these require applied math. And yes, one can most certainly find some mathematical models online that help you do that, but these still require some preexisting knowledge to use. Unfortunately, this is the type of math you don’t learn in a regular high-school curriculum, as we are more focused on solving mathematical equations, not so much on real-time applications. I have a great link for this – it illustrates how much effort Pixar’s team put just to get a perfect reflection in WALL-E’s eye. For those that want to see a bit more about it, I will share here the link here – it’s a great TED talk by the Pixar team.

I must admit I never thought of that, but it’s totally logical. Since you mentioned lots of people study computer science to make apps or video games, what did you want to do with your degree?

Great question with an unfortunately not-so-good answer. I didn’t know at the time. Those first two years passed by, and once I needed to choose what to do next, which subjects to choose, I was totally confused. I mean, we learned the same basics as electrical engineers do, but now out of a sudden we were faced with a challenge of decision what to carry on doing. At that time, I really started asking myself: what does all of this have to do with computer science? What do I really want to learn? Developing apps? How is that even connected to all of the things we’ve studied so far? What nobody tells you is that you need to be proactive, you need to talk to people and ask for their experiences, or you need to get active somewhere else for the sake of finding yourself. And now, after teaching abroad for a few years, I can tell you that other programs are often more project-oriented, even problem-oriented, focusing on giving you the basics of critical thinking and hence setting you up for the next stage in your career.

How did you solve the mystery of what to do next?

As I said, one needs to get proactive. I was semi-bored and decided to do several things at the same time: I was still taking French horn lessons with my high school professor Ivan Ksenek and later with Bostjan Lipovsek at the Academy of Music in Zagreb (both were big influences on my life). I joined NGO Marin Getaldic, which organizes the Young Mathematicians’ Summer Camp and signed up for an internship at Ericcson. From all these I learned that, as much as I appreciated the internship in a company, I realized that was not the type of work I wanted to do. Additionally, as much as I loved teaching students at the Young Mathematicians’ Summer Camp, I never liked competitions. I think there are more fun ways to teach science to kids. Competitions also develop different skills than what you might need for actual research (but all of those skills are useful for researchers as well).

Through organizing scientific popularization lectures for the camp, I met Kristina Majsec who introduced me to the Summer School of Science, which was focused more so on scientific questions. I decided to give it a try. Furthermore, I had the privilege of taking the Intro to AI course by professor Jan Snajder. This was a game-changer and it helped me figure out what I wanted to do. The course had hit all the important points – it was an interesting and multi-disciplinary subject, which nicely complemented my interest in multiple areas, and it was taught by a genuinely enthusiastic professor. I figured that was something cool and decided to point my bachelor’s thesis in this direction. Specifically, I was working on a cellular classification problem from images.

Wow, that must have been a crazy summer or two. But at least you found yourself in a specific subject. Did that help you with your master’s studies?

To some extent, yes. But it was really a combination of all the things previously said. Firstly, I realized that I really wanted to learn more about AI. Secondly, through discussions at S3 I realized that I definitely wanted to do a PhD and hence I started being more strategic – what kind of things can I do during my master’s studies to increase my chance of scoring a great PhD position? And so I calculated the best approach and decided to be proactive again by looking for summer internships, this time more focused on the scientific aspects of AI. I applied to many places and got an internship at the Josef Stefan Institute in Ljubljana, at the Knowledge Technologies lab, under the supervision of Nada Lavrac. This turned out to be such a great decision, as it reinforced my ideas on what science was and how much I saw myself in it.

Did you do your master’s thesis in the same lab?

No, my plan was a bit more strategic. At that time I already knew I wanted to do a PhD. I was lucky enough to be surrounded by people who went through the same process, so I learned that the best way to get into a PhD program is to show that you have the potential for research. The easiest way to do that is a master’s thesis, so I decided to search for an opportunity to do it in a well-known lab. And searching for these labs was a great experience – it was the best way to find out what the pressing research questions were and who the well-known researchers were to accompany them. Long story short, I found several research labs that interested me and it turned out that I could spend a year at two of these universities through the Erasmus exchange program – KTH in Sweden and KU Leuven in Belgium. I was selected to go to Belgium.
The thesis arrangements got pretty complicated – immediately. In Croatia, the last semester consists only of thesis work and you basically create your own topic together with the mentor. At KU Leuven, you spend an entire year working on it and you pick one of the offered topics. Not many Erasmus students do a thesis there, so I was completely outside of the system. No one really knew what to do with me. So I had to figure everything out by myself – find a mentor, think of a topic, figure out the paper I needed to fill in and coordinate everything with my home university.

So how did you choose the mentor in the end? Or did they choose you?

I was a bit lucky and (un)lucky to some extent. The previous summer I did my internship in Slovenia. My mentor in Slovenia was a good friend of one of the professors in Leuven, which is how I got to know his work. So it was easier to prepare myself and set up the stage for my project as something that I would love to do. It was basically a topic that combined my work from Slovenia and my undergraduate thesis. In my mind, I had this nicely set up project and I contacted the professor immediately. He liked the idea and allowed me to do my master’s thesis there.

Sounds good! So why did you say you were unlucky?

Well, the problem was that such an approach directly crashed with how they did things in Leuven. KU Leuven is a research-oriented university and master’s theses are designed with that in mind. You rarely get to make up your own thesis, but rather choose from a set of offered topics. Such master’s theses are the mini-projects of PhDs and postdocs in a lab and are very related to the ongoing research in the lab. In other words, you are expected to do proper research and solve an unsolved problem, at least partially. And as your mentors are personally invested in the topics, they guide and coach you through the entire process. The catch is that with my own topic I missed out on all of that. I mean, the PhD students and professors were helpful, but they knew little about the topic. And back then, I did not know how to ask proper scientific questions nor how to find interesting problems. Looking back, I would have been much better off just asking for the topic to be provided for me. For all the readers that will end up in the same situation – make your life easier and rather ask what type of projects the lab already has.

You mentioned that you needed a core skillset. Can you be more precise?

For starters, I needed to learn how to ask an interesting scientific question. How to formalize a problem and how to approach it, without the exact steps to do it. To give you a contrary example, at FER you get an assignment (problem), but you know exactly what to do to solve it – you just follow the instruction and it will be done. But science doesn’t work like that! We don’t know the solution or the steps needed to reach certain goals. Actually, often we don’t even know what the goal is. One needs to refine the problem as much as possible and take it apart into small solvable parts. It has a lot to do with critical thinking as well, which is something that we don’t practice enough (if we practice it at all).
Studying in Leuven was also a shock academically. Firstly, all the exams were open-book. This means that you can bring all the literature with you. No one cares how much you can memorize. The exam questions were formed in a way that you couldn’t find an answer in a book – you needed to apply your knowledge to a new problem. There were typically a few questions and you get 4 hours. I don’t think I ever left the exam early – there was simply a lot to write. Some courses didn’t even have an exam – you would be working on a big project for the entire semester and, at the end, you would present it to a professor. Secondly, that was the first time I had an oral exam. We never had them in Croatia back then. But everyone assumes you know how it goes. So, I was told the following: we’ll give you the questions and you have 3 hours to sketch your answers. I though that sketching it means to write down my thoughts on how to answer that question. Oh, how wrong I was! I was supposed to write my complete answer – the professor would then read it and ask me the follow-up questions. My sketch was literally just a few bullet points; the professor must have thought I was an idiot. I almost failed that exam. And I studied sooo hard for it! And that’s kind of the Croatian system – learn a lot because you need to know it for the sake of passing it.

Did your grades from FER get affected by this year abroad and limit your chances for a PhD position?

Yes and no. Indeed, I got some barely passing grades (like the previously described oral exam), but for some courses I was among the top students. Overall, I did well, better than in Croatia. But interestingly, my average grade went down significantly when I came back. The reason was that my grades at KU Leuven needed to be mapped to the grades in Croatia. Belgians use the scale 1-20 and Croatian universities have grades 1-5. Even more, the grades at KU Leuven are not designed to be linear: there are 9 levels of failing, 10 is the passing grade, 11-14 are mid grades, and 15 and above are really good grades. Actually, if your average grade is above 13.6, you get a distinction when you graduate. You see the problem – it simply got mapped linearly to the Croatian scale. But it didn’t ruin my chances for a PhD because grades are not everything. Lab heads are more interested in what you can do, your extra curricula, and maybe certain subjects from your studies, but they couldn’t care less if you were not the top student. During my interview, they were mostly interested in my internships, S3 and all the other things I did. I did really well on courses directly relevant to my PhD and that’s all they cared about. And I think this is true for most places abroad, while in Croatia grades are super important. As a matter of fact, the PhD positions often go to people who were top students, without testing if they actually have the other skillset.

From what I know, you continued your PhD in Leuven. Which group was it?

Well, at the end of my first semester in Belgium, I asked my professor to write me a letter of recommendation, so that I could apply to some PhD programs. But he took me by surprise and proposed to discuss the PhD positions in his group. He told me that, from his experience with me, he thought I would make a good PhD student due to how I worked. In the end, I applied for several other positions, but decided Leuven matched my interests best. Learning from my master’s thesis experience, this time I picked a project similar to the interests of my mentor. Interestingly, it was a topic similar to the courses I hated the most during my undergraduate studies, the ones I was bored to death by! I was simply taught about it the wrong way. Now, being in a setup which nurtures learning and mentoring, I’ve relearnt how interesting that topic is. And six years later, I still find it just as fascinating.

Sounds like an interesting project. Can you tell us about it?

AI is all about mimicking human intelligence and problem-solving skills. (And it’s still bad at it.) What attracted my attention is how humans can solve problems at different levels of abstraction. I decided to try to mimic that. Let me give you an example. Let’s imagine you’re planning a trip. If you want to go to Leuven from my home in Pozega, you will not simply get into a car and drive. Instead, you will first find a flight from Croatia to Belgium; this brings you to the correct countries. Then, you will realize that the airport in Croatia is in Zagreb, not in Pozega, and will be looking for the way to reach the correct city. Similarly for Belgium – the airport is in Charleroi, not in Leuven. Once you figure that out, you will be thinking how to navigate the cities until you reach the house you need.

These are three levels of abstraction: country-to-country, city-to-city, street-to-street. This is a uniquely human ability, no other animal can do this as far as we know. Actually, cognitive scientists believe that the ability to think abstractly is one of the reasons why human intelligence skyrocketed, while apes did not follow. And there are countless examples of how easily we do it, while computers struggle. In our work, we managed to teach a computer to use abstract thinking in quite simple problems, but as soon as it reaches a more complex one, it fails miserably.

Since you are still in the same lab, now in a postdoc role, are you still working on the same project?

In broader terms yes, we are working in the same field, but there is a stark contrast between a PhD and a postdoc. During my PhD I was focused on a specific problem, a specific project, as it allowed me to develop myself as a scientist. However, this specific problem during my postdoc became an array of different projects. The role of a postdoc is to bridge time between doing a PhD and becoming a professor, which involves not only working on your project, but developing a long-term scientific vision, designing multiple projects and mentoring students who work on these projects. Actually, the time I spent solely working on a project is much shorter than before because, in addition to everything we previously mentioned, I also collaborate on different projects, have to write grants to get extra funds, organize workshops on conferences, organize teaching etc.

But as a person who loves science so much, don’t you miss doing more scientific stuff?

It depends. This way of working actually suits me. (Though I would frequently complain that I want more time for research.) I like to think about problems and formulate scientific questions, figure out solutions and even design experiments. But real work assumes implementing solutions to those, which I am more than happy to give away to my students. I am okay not getting my hands dirty with coding, which is fun in the beginning but full-time troubleshooting in the end. Furthermore, I love to bounce ideas off my colleagues and then develop new ones, which is super cool. But this role allows me to work on more projects simultaneously – during the PhD, I used to work on a single problem, whereas now I simultaneously work on three or four. This works much better for me because I can get easily distracted by a new idea and become frustrated when I have to wait to work on it. The only thing I don’t like, or phrase it better, what I didn’t like in the beginning was writing grants. But I’ve learned that it’s almost like bouncing off ideas and thinking about scientific problems in the long-term. And this is what I like to do. One just needs to write them down, at least what they might need and how they might develop over time, without putting too much effort into details. What you deliver must be clear, but it can be a rather simple part of a bigger problem. Once you get a grant, then you can start expanding your initial idea.

What is next for you?

Though I’m now safe for a while with my personal postdoc grant, I started slowly applying to different professorship positions. I am doing a bit of research prior to applying, where I am trying to learn how much a particular university invests in research vs. applications, how much dedicated teaching time there is etc. I don’t really care about the country, although my tendency is towards more developed European countries which are harboring some excellent universities.

Summer School of Science

As I mentioned, Kristina Majsec gave a lecture at the Young Mathematicians’ Summer Camp and told me about S3. Technically, she invited me to organize a swapshop in 2011. I did not end up doing it in the end, but she was impressed that I bothered to find my own substitute and invited me to be a project leader next year.

That’s nice! So, how did you like it in 2012?

I think one of the most important factors was that S3 was something I missed while I was in high school! At that time, it would just be hard to get information about the School if nobody told you. And that feeling of being surrounded by like-minded individuals was amazing. I remember that it was challenging to prepare my project at first, but once I got into it, it was easy. My students were also extremely interested and asked lots of questions. Most importantly, I learned that, if I get interested in something, I don’t mind doing boring stuff for it. For example, when I organized the School in 2013, I didn’t really enjoy doing the fundraising and all that stuff, but I learned to like it for the sake of a greater goal.

• 

You also joined the Society of Out-of-Frame Education (EVO) in 2012, the NGO that organizes S3, and today you are the head of it. What are the challenges of that position?

Yes, at that time there was a rule that one needs to join the NGO for the sake of organizing the School, and since we expanded to Pozega, it was a clear thing to do. But being in this position is a completely different pair of shoes. It feels more like politics and is tricky to get things done. And it was, and it still is an interesting experience, working with volunteers on all different kinds of challenges (mostly administrative ones). What is so challenging about it is that it is an entirely volunteering-based project. People dedicate their free time to this for a greater cause. And of course they want to do stuff they enjoy doing, like designing projects, finding people for the Summer School and such. But there are many more boring things that simply need to be done if we want to have S3 and grow our activities. Such things are typically boring, like fundraising, negotiating with the hotel and the restaurants, maintaining good relationships with everyone and so on. These can feel like very lonely jobs and usually very few people volunteer to do them. But it is completely understandable that people don’t want to spend their free time on boring stuff.

• 

• 

We need to make that cooler then, no?

That would be great, I just don’t know how to do that (laughs). I think we need to be aware of all of the things in the shadows and make our peace that they simply need to be done. Everything that makes S3 amazing requires a lot of work and that work is often tedious and uninteresting. You just don’t see everything that the organizers need to do. And while many of those things aren’t exactly the most interesting, you will learn a lot and develop new skills that you will need sooner or later in your career. We should also make clear what the expectations are. I think many young people are too harsh on themselves, like if they apply for funding and then they don’t manage to get it, they feel like they failed everyone! And that’s totally not the case. The effort is appreciated in any case. Since 2013, I cannot tell you how many times we failed to get extra funding! But each time we learn something new, reiterate it and get better at it. So don’t be afraid!

If someone is interested, what should one do to become part of EVO? What kinds of things can they do there?

You can just write to us via email and tell us you are interested to join. I would tell you to check our website for more information on what we are currently doing, but the vast majority you won’t find there. Yet. But, we need help in many aspects: writing grants, creating new projects, joining the old ones, helping updating our websites, working with alumni, etc. The amount of tasks is growing rather than declining as we often get new ideas on how to make things better.

I totally agree with you, there is so much to do! You can even join us here, on the blog and write different types of stories or have interviews with previous alumni. It’s always a blast!
Before we finish up, I am curious, what does your free time look like?

There is less and less of it lately, but I do enjoy two things in particular. I really love to cook. That started out of necessity as I sometimes didn’t have time to go to the student restaurant (or just didn’t have the will), but then I discovered that I enjoyed doing it. I love to eat, try out new recipes, and discover new cuisines. I will never say no to going to a Thai, sushi or ramen restaurant! Furthermore, maybe not so exciting, but I like to read. I was always an avid reader and, as I am quite the early bird, I actually read for 2-3 hours before going to work. Sometimes that boils down to reading papers, but also books. According to my broad interests, I would read books about many different topics, most often non-fiction: human history and evolution, cognitive sciences, social issues, and what not.

Lastly, what kind of advice would you love to leave our audience with?

There are a few things I wish I knew sooner. Get rid of the fear of failure. The sooner you get over it, the faster you will learn new things. Most of the time (and I really mean most of it), nothing bad will happen if you fail. You will actually learn more from failures than from successes. If there is an easy problem to solve and you solve it, is it really so interesting? Challenge yourself with things you don’t understand and you will develop skills along the way. If you are not certain whether you would like a certain internship or studying at a specific university, just go for it. The point is that you don’t know and in the worst case you will reduce some uncertainty. I did many of those things, and I cannot remember a single one that I regret. It was always a worthy experience.

Also, stemming from my own experiences – do not believe your own expectations, but rather take the opportunities that occur along the way! My own expectations often tricked me – things I expected to be boring were super interesting. I mean, look at me – nowadays I am working on super interesting topics that I was bored with during my studies. Just don’t waste any opportunities that arise, because you don’t know what they are exactly, take a chance and have fun!

Favor breadth and be selective about depth. I think science is becoming more and more specialized and I don’t think that is a good thing. I can only really speak for AI, but people are so specialized nowadays that they keep reinventing stuff. That is not how things in science should be. Having a broad view makes it easier to see big problems and I do believe it also makes you a more competent scientist.

Thanks, Sebastijan, for a nice interview. For all of you still full of questions, feel free to leave your comments below!