Master Thesis #1

Second try, second chance! Five years ago, I have written my first master thesis. An Egyptological study about how the Ancient Egyptians used graphical tools and layout to structure their papyri. And for some reason I wonder why I have never published any of my results on my webpage. I definitely should have done that! But that’s for a second project.

This time, I have changed the university and the basic subject, as this master thesis will be done in computer science. I am studying CS with the minor “machine intelligence” – there are actually three chairs in Basel dealing with this topic, leaning towards artificial intelligence, biomedical data analysis, and computer graphics and vision.

Now, the first and obvious choice for me would have been the last one – I have studied media informatics in Munich and I do work a lot with 3D models in my day job. And it is a very fascinating field of study! I had some courses there and they were magnificent. The same with the biomedical data analysis, I had a full course on bioinformatics and really loved the algorithms and how they were used to solve problems which otherwise might take ages to calculate. Buzzword “sequence alignment”.

For some reason, though, artificial intelligence kind of made it into my head. For several reasons, to be honest. First, planning and the optimization of plans is something which has been stuck in my head for almost 30 years. As a child I went along and tried to be as structured and reasonable as possible: where am I? Where do I want to go? Which obstacles are in my way? Which options do I have? Now go find a route through this maze. A way of thinking of problems which can be adapted to anything in life. Trying to find the best or cheapest path from my current state to a goal state, that’s just how my brain worked.

Second, I love how this field has been very formal, very focused on maths during my studies. Don’t get me wrong, all informatics are about maths. You cannot do advanced informatics without knowing your maths. But in this particular field I learned a lot about how to define things formally and properly, how to write those things down and how to prove all that. It is – I have to admit it – quite a challenge. But it is well structured and very logical. Quite a contrast to my first master thesis, done in humanities, and quite a contrast to my bachelor thesis where I had to program, but not really to do anything mathematical.

Which leads to my third aspect – the challenge. Yes, it is a challenge, and yes, I am full aware of that. I could make my life much easier here. I don’t want to. I kind of need this challenge to get a deeper understanding of my own abilities. I want to convince myself that I can do that – and not just fulfill the task, but do it in an excellent way. This may sound arrogant, yes. In fact, it is not. To the contrary. I abuse this work to make myself more comfortable with my skills. And increase them. As I usually say: reach for the utmost to extend your possibilities. You won’t grow in life if you stick to what you already know.

This week I had my first meeting with my potential supervisors, and they spent some time on me and presented me with two topics I could choose from. Both topics sound very interesting, but challenging – I guess you will understand I don’t write them down here without having read the first papers. Better not write something down which is complete and utter nonsense. 😉 But they don’t sound like completely undoable jobs. Which is an important first trait.

One good thing about doing a master at my new university: the master thesis is kind of split into two parts. First, you have to do a preparation phase for your master. You choose a topic and start reading on it, gather all the relevant literature, figure out the current problems and the basic history of the work in this field. After one month, you report on your findings and either pass or fail. If you pass (and have finished the rest of your studies in a good enough way), you can finally start your real master thesis, a work to be done in six months.

The advantage, compared to my first master thesis: a lot of work is done before the actual work starts. Not having to get a first grip on the topic, already being familiar with the fundamental literature, having had some first thoughts about the problem and potential solutions, that’s really a big help when starting into a project. And it doesn’t take additional time. Besides, if you figure out after this one month that the topic does not really fit, you can still change. Better now than when writing your thesis, right?

Anyway. I am really motivated to do this in a good, motivated, organised and concise style. I have learned a lot about complicated project works over the last couple years, and I have improved my own discipline and tactics as well. Let’s see if it works out. And it might be a nice thing to leave some notes here on my website from time to time – not as a general “blog” of my master thesis, but to let you and myself know the current state and my thoughts about the work.

So, I’d say – let’s get started! 🙂

Machine Learning and Politics

Believe it or not: machine learning, a computational discipline to make our beloved little computers more “intelligent”, actually teaches me some lessons about politics. These are no super new insights, but I sincerely love how I was able to transport my knowledge about machines to my experience of daily politics – and thus wanted to share my thoughts with you.

When machines try to learn, they always need to figure out some model of their respective environment. A robot, for instance, may have to derive a model how to get from one point A to another point B without using too much energy. Which movements are costly? Which ones are not? Where to go to reach the goal at all?

Another robot may stand in front of a box with a lot of sandwiches, handing them to fascinated students. Its task: predict the next sandwich flavour. Will it be turkey? Or cheese? Or just pure cucumber?

Well, you may find better examples of what robots can do, it doesn’t matter at all. In all those cases, the basic principle is the same: we have a bunch of hypotheses about our environment and want to figure out the most probable one. To do so, we use two major ingredients: prior (expert) knowledge about our hypotheses and observations, which may update our knowledge.

Actually, politics are just the same. The more you deal with or read about our daily politics, the more you learn. You gain some expert knowledge, which allows you to predict whether some events are more likely to happen or not. At the same time, you watch our politicians, given our global environment, and observe how they react, what ideas they produce, which strategies they apply to convince people. The political stage is the environment, the politicians are the actors, and their actions are observations.

In a grown-up political system, there is little which might surprise us. Think of ongoing elections: is it surprising that politicians of all parties claim themselves to have the only possible solution while all the others are horrible losers? Are we shocked that politicians transform the biggest loss into a magnificent victory after an election night? Are we surprised that politicians constantly repeat the very same sentences, as if they would become reality afterwards? Not really. This matches perfectly with our expectations.

There are some elements, though, which don’t fit. Think of the American President Donald Trump, for example. Nobody, me included, believed in his victory. My prior knowledge said: the Americans cannot vote such a man. They proved me wrong. Another example: Brexit. I was sure the British people could not be so stupid to leave the European Union. They were. Surprise, surprise.

These events, these results happened, because we all – including our politicians – were so sure about what might happen. Our prior knowledge has been confirmed and confirmed multiple times, and thus our prediction was super clear. However, reality sometimes doesn’t fit the model. We got surprised. Wow.

The result: the certainty of our prior knowledge went down drastically. Think of the elections in the Netherlands. People were quite sure that Geert Wilders might lead his party towards victory. Polls were ignored, said to be wrong all the time. In the end, they were right, and Wilders lost – a little healing confirmation to our wounded prior knowledge.

Next stop: France. The closer we are, the more worried we become. Our prior knowledge tells us that Marine Le Pen has no chance at all – but does reality know that? Will the French people actually vote against the European Union, the European friendship? We don’t know yet. Dozens of articles are written, some of them saying “Why Marine Le Pen will never win!”, others “Why Marine Le Pen could actually win!”. Yeah. Whatever.

We see: our observations of reality have drastically changed our prediction of the future. The once so clear posterior probabilities were reduced, new hypotheses joined the game. Suddenly, there is movement, there is a change in data.

And at the same time, people start to become political again. They see that things are moving around, and they either want to throw it over – or they want to rescue the system. They were inactive as long as the machine was running smoothly. Now it doesn’t. Now they interfere.

Politics are the environment, and we are the machines. The agents. We constantly update our knowledge based on what we observe – the daily news, political statements, our own experience in the streets of the cities. There is a little twist, though: we are part of the environment. We interfere with it. Our data is not independent. Which makes things super complicated.