'What Matters Is the Way of Thinking, Not the Field'Interview with Hussein Mohammed

18 February 2022

For Hussein Mohammed, the differences between computer science and the humanities are much less significant than the similarities. In this interview, he talks about the winding path that led him from teaching black holes to young refugees in Syria to researching pattern detection at CSMC.

Hussein Mohammed, you are a computer scientist who is surrounded by people who spend their days scrutinising manuscripts. To an outsider, this might seem as if two different worlds are colliding. So tell me, how did you end up here?  

Well, I joined CSMC after doing an MA degree in Portugal, where I wrote a dissertation on ‘Object detection in complex scenes’. The detection process involves recognising, classifying, and localising certain objects in images, for example a pen. Look… (He picks up a pen and puts it on an empty table.) This is not a complex scene. If this was an image, it would be very easy to detect the pen. But now look at this. (He lifts the pen and holds it diagonally in front of a colourful book cover.) Now it’s a complex scene: the contours of the object are no longer easily recognisable and the background is not uniform. I was trying to develop a system that could detect shapes in complex scenes using just their outside contours without the extra information provided by colour and texture, for example.

So far, this has nothing to do with handwriting, though.

But the idea can be applied to it. In handwriting analysis, you often want to know whether two manuscripts were produced by the same hand, for instance. To do that, you could ignore things like the thickness of the line or the colour of the ink and characterise the styles just by looking at the outside contours and the shapes they create. This is how I initially got involved with CSMC: during my PhD, I tried to analyse handwriting in images and measure the similarity to handwriting in other images. One can eventually generate similarity measurements which can indicate that the handwriting in a certain set of images looks much more similar than the ones in another set.

What about your current research? Are you still working on similar problems?

The main topic of my research at the Cluster is not about handwriting styles, but pattern detection. Let’s assume you have some digitised manuscripts, maybe a hundred, maybe a million. Now imagine that you want to search for a specific seal, signature, or drawing, and you want to know which images it occurs in. In more abstract terms, you want to detect a visual pattern. This would be a huge waste of time if you had to do it manually. Theoretically, this can be automated. But most of the current state-of-the-art methods depend on the existence of annotated training data. The approach we developed at CSMC does not rely on the availability of annotations and yet it has achieved state-of-the-art results. At the same time, it is general enough to detect words, seals, drawings and so on. 

That sounds brilliant, but what is so bad about using annotations?

By ‘annotation’, I mean the x and y coordinates of the occurrence of a pattern, the width and height of the region, and the label. Depending on the number of images, creating annotations can take a very long time. What’s more, in many cases, you can’t just leave this job to anybody; it requires expert knowledge. But experts have much more important things to do than spending their time annotating images.

Moreover, the current state-of-the-art detection methods depend on the availability of lots of annotations, sometimes even millions. For the vast majority of research questions here at the Centre, you don’t have these amounts. All you have may be some digitised artefacts for your private use, which you cannot even upload to a cloud service. This is why the researchers here at the Centre and scholars in the humanities in general need a method that does not require annotations.  

I started to see how scientific thinking can be an indirect key to solving problems in my society.

Before we return to your research, let’s go back a bit and talk about the path that ultimately led you here. Prior to becoming a researcher, you worked for several NGOs, including the International Refugee Committee. What did you do there and how did you find your way from that very hands-on activity to the more abstract world of academia?

I always had a strong interest in science. But going into research was not a feasible option in the countries I grew up in. I was an undergraduate at Baghdad University when the war started in 2003. To finish my degree, I had to move to Jordan and then to Syria, but I didn’t have a working permit when I graduated. So, I decided to work as a volunteer in the education sector to help refugees from Iraq. During my work there, I started to see how scientific thinking can be an indirect key to solving some of the problems in my own society.

Isn’t scientific thinking a subordinate issue when you are facing people who have just escaped a war zone?

The moment you realise that what you believe is just an opinion and not necessarily based on any facts, the moment you realise that facts matter, that they can only be validated by evidence and experiments – once you embrace this way of thinking, a lot of myths and stereotypes will fall apart. This is key to questioning some traditions and local templates that could constrain societies. I tried to pick scientific topics that seem interesting to young teenagers, like black holes. I asked them about certain preconceptions they held about the universe and gave them evidence that what they had believed for a long time is actually wrong – just to show them that it is okay.  

…to be wrong?

To live with the feeling that you were wrong for a long time. This is really hurtful. Suddenly, the image you created about yourself and the world is distorted. It’s scary. Many people find it easier to create excuses just to keep their image of the world as it is.

If that is the thought that motivates you, I wonder why you ended up in computer science rather than physics or philosophy.

When I was working in Baghdad, I got a scholarship to do my MA in Portugal in a robotics lab. With every month that passed and every book I read, I grew more and more convinced that all kinds of scientific research have a lot of things in common – whatever the field. The way we analyse problems and look for evidence, the way we design experiments to support or contradict our assumptions, the way we don’t take our assumptions about the world personally and are willing to change them anytime – in my opinion, these are fundamental principles in science. That is why I didn’t find it hard to choose between different fields. Initially, I got drawn into science by my interest in physics and eventually I became a computer scientist. That is not a big deal for me, though. What matters is the way of thinking, not the field itself.

Computer scientists and scholars from the humanities can be amazingly helpful to each other.

That sounds like a fruitful attitude, given your position at CSMC. So, let’s come back to your work here. Your starting point, I suppose, is always to identify areas of common interest between yourself and the scholars in the humanities you encounter here.

Yes. The workshop on Computational Palaeography I co-organised at the ICDAR 2021 conference in Lausanne is a good example. By definition, the goal of the workshop was to bridge the gap between computer science and the humanities. This gap is hard to overlook. At the same time, both fields can be amazingly helpful to each other. Computer scientists need to see that many research questions from the humanities are very interesting, deep, and challenging. And researchers from the humanities need to realise that they can make a lot of headway if they understand some basic concepts in computer science.

What does that mean in practice, also compared to the kind of work that computer scientists normally do?

A typical goal for a computer scientist in my field of research is to achieve better performance on standard datasets. These datasets are used in computer science in order to compare results from different researchers. The state-of-the-art results on these datasets are improving rapidly. To get published, you are constantly running against the clock. In contrast, I am dealing with unique datasets to solve specific research problems in direct collaboration with scholars in the humanities. Methods developed for standard datasets are rarely applicable in these contexts.

Is that the reason why you can afford to work on your own? Because you don’t have to be faster than other people working with the same data?

It’s true to an extent that I am not directly competing with anyone. Eventually, though, I need to publish my work for the sake of my career in computer science. In my field, one of the main criteria for getting your paper accepted is to have some kind of comparison: how well does your method perform on standard data sets compared to other state-of-the-art methods? Therefore, an additional barrier I need to cross is to find standard datasets and tasks, which are somehow relevant to my actual research problem, and then produce competitive results with them. So I have to address two tasks: tackling actual research questions on actual research data, and then applying my methods to other standard datasets in order to publish in my field.

You have to cope with the worst of both worlds, then: time pressure in computer science and the intricacy of research questions in the humanities. What keeps you motivated?

For me, achieving slightly better results on artificial datasets is not fulfilling; it would not have an impact on anyone’s life. In my work at CSMC, there are actual people trying to solve actual problems from manuscript research. If I get a method to work well, I can help them. I can be part of the solution. I find the pure feeling of being helpful extremely satisfying. I love solving problems, and I love it even more when other people find my solutions to these problems useful. I guess I’m still driven by the same emotions as a child is.