5

I'm a new PhD student studying AI and I've never attended seminars before. When I attend the seminars held within my lab (a lab member presents a paper) now, I have a lot of questions about the details of the theory. But I rarely see anyone asking about the gory details of theory. Do I have the wrong approach here? the approach being, going with a mindset of learning what the paper being presented did and how they derived everything and not thinking about what it's useful for and how their contribution can be applied?

For example, if someone shows a loss function I've never seen before I instantly think of the notation, the symbols and what they might possibly mean (in a graphical sense mostly). But the presenter always rushes over these mathematical equations and derivations alike.

Another way to ask my question is- Should I be thinking of the bigger picture of a new AI solution (say, what the input is, what the output is and how its evaluated) or should I continue to think about the smaller picture (the mathematical formulation and derivations) first and only then the bigger picture.

To extend this question to seminars of any field, my question is- What is the correct attitude to assume when attending a seminar? (this encompasses thinking about the paper, how to ask questions properly and other factors I might not have thought about)

7
  • I feel like there are a mix of different questions in here, it would be best to focus on just one of them. Also, what do you mean by a "seminar", as there are many types.
    – Bryan Krause
    Commented Apr 30 at 19:41
  • By `seminar' I mean when a member of the lab presents a work that is interesting to them. PS. I didn't know there were different types of seminars. Commented Apr 30 at 19:53
  • This is a lab you're in? Yeah, a lot of things get called "seminar"; some examples in my personal experience would be: 1) A talk by an invited speaker or student, 2) A short standalone course usually not part of a degree program that you pay to attend to learn something, 3) like 2, but part of a conference, 4) a credit bearing course as part of a degree program that focuses on review of academic literature rather than a textbook with lectures. There are additional meanings outside the US.
    – Bryan Krause
    Commented Apr 30 at 20:10
  • Yes this is within the lab I'm in. Thank you for the outline of the different seminars. Commented Apr 30 at 20:16
  • This is far too narrow for us to answer here. Groups have different norms. Commented Apr 30 at 20:38

2 Answers 2

3

The answer by @Bryan Krause is already good and for sure it makes sense to ask your advisor about this, but here's my take.

Your question is an interesting and complex one, and it reflects well on you that you ask it.

The question has several aspects, in particular behaviour in the seminar vs. how you should think.

In principle both the "smaller" and the "bigger picture" are important. Care for the details is a good thing as important things can go wrong without it. But it's also important to not get lost too much in the details so that you lose the bigger picture from your view. To what extent you need to focus first on the "smaller picture" depends to some extent on what kind of person you are, how you work, and how you think. If you can, really try to care for both.

In particular, the fact that currently there isn't any discussion in these seminars on such details doesn't mean that you are wrong thinking about them. Often there is no agreed etiquette for seminars like these, and how they go is a product of social processes. Many people just adapt to what seems to be the dominating style without much thinking, and if the first seminars went rather randomly in a certain way, it may well happen that then all later seminars are like this. Also many students are shy and are worried to say something wrong or to reveal that they lack some fundamental knowledge if they ask, and so they won't.

Such social processes can be changed, and will change to some extent automatically if somebody comes in and behaves differently. In principle I imagine (of course I don't know the background and "history" and the people involved) that you can just ask what you're interested in even if nobody else does it, and some may well be happy that you do that and may even be inspired and start asking stuff as well.

But be careful; it is possible to overdo it. You will find out later yourself that time is never enough to present everything that is interesting and to explain everything in full detail. If currently there is a habit of not having questions, presenters will be "optimistic" about how much material they can present, and they will prepare for it. Many presenters are grateful for questions (I certainly am), but a problem can be that a presenter is eager to go through all the prepared material, or at least to discuss some later parts, which they may not reach if too much time goes into questions earlier. Normally I expect a presenter to leave a certain amount of time for questions, so I expect that you can well ask a question here and there, but people may not be happy if you take a considerable percentage of somebody else's seminar time for your questions, particularly if they are not used to this happening.

Many questions that you have in mind may probably be legitimate and good, but as there is so much to say, there will not be space for everything that is potentially of interest.

So I'd encourage you to start asking one or a few things as a trial. Consciously only take a limited amount of the seminar time of another presenter. See how it goes and adapt, depending on experience. You can bring in your style and thoughts; this should be enriching for the group, but be aware of the dangers.

7

Since this is just a seminar in your lab, why not ask? Not as part of the seminar, just in your informal conversations with the people you work with. These are people you should be conversational among. You don't even have to ask it as a question of etiquette, you can start with just the observation you have: that people don't talk in depth about the "theory" in the way you see it. If you want to know what you should be focusing on, ask your advisor. I think the questions that other people ask are a good hint of what they are focused on but not necessarily what you should be focused on.

It's possible that some of the things you're questioning are considered apparent to people who are not as novice, and maybe as you pick things up you won't have those questions anymore. Alternatively, maybe those details go right over the heads of the people talking and they don't actually know how any of it works. Both of these patterns are familiar to me in the impressions I have of the field you are in: mathematicians like to skip steps they find trivial, and AI/ML practitioners often get training and expertise in tools rather than actual mathematics.

In my experience, a seminar like this is mostly for the benefit of the speaker, to let them practice presenting their work and also to encounter the questions and challenges they'd receive when presenting to an outside audience. As such, it's not usually a time to educate other people in the group, but it varies; sometimes it's worth it to everyone to go into a deep dive about a particular decision and sometimes you may find that the person doing the work didn't even think about the alternatives or implications.

I'll add that sometimes having a new person in a group is a great way for a lab to identify patterns they've fallen into and to break out of them, and academia really benefits from this since there's a constant rotation of new students in that have had exposure to different things. I wouldn't push too hard on changes as a new person in any group, but that doesn't mean you can't ask questions and add your perspective, just be prepared to learn from the responses you get.

Not the answer you're looking for? Browse other questions tagged .