It’s over, yet again. Always hard to believe how tough a term gets in the last couple of weeks, and how long afterwards the subconscious refuses to acknowledge that there’s no longer any reason to be tense. The last exam was yesterday morning (today is Saturday), and even after 8 hours in the pub and a decent night’s sleep I am still not nearly completely unwound.
In the end I did prepare some more for adaptive systems. I rehearsed the presentation a few days, particularly since I had a hunch it might be too long (it was). Also I wasn’t yet decided whether I should talk English or German. The written paper and the presentation slides are in English, which I always prefer if the sources and the technical terms are in that language. I rehearsed both versions, but in the end decided to do the talk in German. The professor has a reputation for being critical, especially when complex information is not communicated precisely, and I thought it wiser to remove any impediment to his understanding me. Which was a good idea, because he said just that in the exam (that I had better talk in German).
The three hours before my exam around 11 a.m. I spent in the cafeteria with two of my co-student who were examined before and after me, respectively. I even worked on my presentation some more, drawing a colored flow diagram to explain the core update algorithm, which also was a good idea, because things like that are hard to comprehend when they’re presented solely orally. Actually I was in a relaxed mood then.
The three of us decided to attend each other’s exams. They were technically public, but the decision whether the public should be allowed in each case rested with the candidates. I suppose we thought it would be a good idea to see some sympathetic faces, rather than be alone in the room with the professor and the minute taker.
I had the feeling our presentations went all well, and the atmosphere was apparently quite relaxed. The first of us to present got an A- with which he was very happy. My talk also went well, particularly since I skipped some parts on the fly and accomplished a precision landing 15 seconds shy of the allotted 20 minutes. The discussion was animated, but friendly, and after just a minute of deliberation I was recalled to the room. The professor told me they had virtually had no choice but give me an A+, and in fact that had been clear to them before I even started to talk, on the strength of my written paper alone, whose scientific format, complete with notes and lengthy bibliography, had convinced them. Truth told, this was an easy victory, because after 20 years of professional experience as an academic, writing a paper with a lot of footnotes comes as natural to me as breathing. I had not really expected that alone would count for so much.
However, after the third of us had presented, the professor and the minute taker took over 20 minutes to discuss his grade while we waited outside. And then he got a straight C. The reasoning of the professor was that he had just played around with the parameters of reinforcement learning and done little else. Yet while I certainly overdid it (yet again) I too had just gone through a lot of trial and error with the parameters of RL and neural networks (well, I did have a neural network). And the presentations, in my opinion, were all pretty good. In fact, the guy who got a C had a lot more numbers and graphs than I had had.
So when we went to the pub as always, the mood was a bit dampened. In fact, out of eight of us who had taken the adaptive systems exam, no less than four had received a seriously disappointing grade. One of our study group (who by his own admission, however, had invested very little time and written the paper in the small hours of the night before the deadline) got a D+. Worse, another student who usually gets very good grades and who had invested quite some effort in her adaptive systems project had been given the choice between a C- and a failing grade (so she could re-take the exam next term in hope of a better grade). And the very last one of us to present had failed entirely.
The problem is that it had never become clear to us what the professor expected. True, he had given us an grading scale in which the conditions for each grade were elaborately defined. For instance, for an A you had to gain insights into “non-obvious complex properties of machine-learning concepts … sufficiently abstract and sufficiently generalized” as well as model your environment “correctly, suitably, and effectively”. Which sounds all very nice and precise, but in the end it still didn’t help us any to figure out what the professor wanted us to do, and how much of it. What might be “non-obvious” to me might well be quite obvious to him, what’s “sufficient” is very much in the eye of the beholder, and so on. The failure of about half of us to get a good grade in spite of quite some effort probably had a lot to do with this insecurity. Then again, three others received an A, two of them without having obviously invested a lot more, or done anything evidently different, than those who got a C or D. And curiously one got a B in spite of doing nothing else but taking an old bachelor thesis and implementing yet again what the bachelor candidate had done–and the professor even knew that! So we are left with the impression that the grades had been, to say the least, a bit subjective.
The mood in the pub also suffered a bit from the fact that, unlike in the three previous terms, we hadn’t all had one final exam together just before going there. The last written exam we had in common this term, computer networks, had been on Monday, and the oral ones in software engineering had been spread over three days, Tuesday through Thursday. Only those taking the compulsory choice module on adaptive systems still had another exam after software engineering, and most of them already on Thursday. So most had been done well before Friday, which made the pub gathering curiously deflated and accidental. We were a lot less than in earlier terms, some came late, some left early. And I personally was and am profoundly tired after this last couple of weeks. Somehow the exams on top of all the excitement over the different job offers and decisions had taken their toll. And besides, I am suffering from some very late bout of hay fever, which usually comes in early June, but with the wet and chilly summer seems to have been delayed six weeks or so.
So I will now have the privilege of taking several days of getting my distance from computer science before we will leave for our three-week vacation in Normandy a week hence. Although I also have to come up with an idea for a presentation in the seminar next term. I am going to give a talk on functional programming in Scala, but on what exactly I don’t know yet. I am just certain that FP is cool and that for a long time I wanted to learn Scala, so here we go.