Coffee Talk Tips

Giving a good 10-minute talk about a complex topic is hard (it is very hard actually), because you can only cover the gist of the paper. The coffee talks are a good way to practice your skills in giving short talks. The coffee talks also help everyone in the lab gets to see a lot of recent work in the field (without having to read tons of papers).

There is no single best way to give a great coffee talk. However, below we provide six tips you may want to take into account when preparing for a coffee talk:

1. Don’t talk about a paper or idea you don’t understand. If you don’t understand a paper after studying it for an hour or more, how can you expect your audience to understand it in 10 minutes? If a paper is too complex, just pick a simpler paper to present and save the complex paper for some other time (when you had more time to digest it).
2. Give a proper introduction to the problem. If your audience does not understand the problem the paper is trying to solve, then the audience will surely not understand (and not even care about) the solution to the problem that your paper proposes. Please keep in mind that your audience may not be nearly as familiar with the topic as you, so you may need to explain details that you think are completely trivial.
3. Present just the key idea of the paper. Typically, a single new idea forms the basis of the paper. You may not even have to read the full paper to get that idea: in good papers, the key idea is already presented in the introduction. If the paper presents many new ideas (this is hardly ever the case!), just pick one idea and cover that one: it is always better to explain one idea well than to explain five ideas poorly!
4. Get rid off the math. The goal of a coffee talk is not to show off your amazing math skills! Math is useful when you want to understand all the nitty-gritty details of a model; but if you need to spend half an hour understanding all symbols in an equation, your audience is never going to understand that equation in just a few minutes. Rather than filling your slides with math, give an intuitive explanation of what that math is saying (that is, give the gist of the math).
5. Do not present all experiments. The fact that a paper presents experiments on 58 data sets, doesn’t mean you should present all of them: nobody likes to read a big table with the performance of 23 algorithms on 58 data sets. Instead, give an intuitive overview of the experimental results, focusing on experimental results that highlight interesting aspects of the model you’re talking about. (When does the algorithm work better and when doesn’t it?)
6. Limit the number of slides. A good rule-of-thumb is that you need about two minutes to present a single slide, which means your talk should not have more than five slides. It goes without saying that you cannot limit the number of slides by putting more stuff on them! :-)

As an example, you could use the following five slides: (1) a slide explaining the problem, (2) a slide explaining the typical baseline approach, (3) a slide explaining the gist of the new model, (4) a slide with highlights of the experimental results, and (5) a slide giving some thoughts for discussion, ideas for future work or applications of the presented work, etc.

You can use the Coffee Talk Archive to see whether or not already a talk has been given (recently) about papers. Note that it is OK to give a talk on a paper that has already been discussed before, but clearly disclose this at the beginning of your talk.

A final piece of advice for the audience of a coffee talk: Irrespective of whether a talk is related to your own work, a good researcher should be able to show an interest and understand topics outside of their research area. This can only help to strengthen your knowledge base. You never know, a coffee talk on a really distant topic to yours might actually inspire your own research in surprising ways. And remember, there are no stupid questions! All questions help us to learn together.