How to Make a Conference Talk

June 7, 2014

Conference talks are a particularly rewarding part of grad school. And compared to reading your paper, you have a better shot at getting people on your side when they see your talk. But because they’re so visible and critical to dissemination, they’re an enormous responsibility to heap onto students—especially if it’s your first talk.

Here is what I have learned so far about the process of preparing a talk.

Leave a tremendous amout of time.

Making a good 20- or 30-minute talk takes a huge amount of time—not just your time, but the time of your advisors, co-authors, and other willing practice subjects too. I know it’s cliché, but start as early as possible. Three weeks is a comfortable minimum.

Start with an outline.

The first step is to make a high-level outline of the talk’s contents. Your instinct will be to jump right in with the slides, but do not do this. Starting with slides will bog you down in details that don’t matter. If you’re like me, you will probably get discouraged at the interminable road in front of you and give up.

To make the outline, it can help to start with the rough outline of your paper: read off the section and subsection headings and call that an outline.

A typical conference talk outline should contain, in order:

  1. An immediate preview (“executive summary”) of the overall message of the talk. This should succinctly, in a slide or less, capture the story you want to tell. It is your chance to explain why the audience should momentarily stop checking their email.
  2. The introduction, where you motivate the problem and describe why your angle on it is interesting and important.
  3. The “body” sections, where you explain your technique. Pay careful attention to the level of detail: explain only the most surprising and distillable aspects of your work.
  4. The evaluation, implementation, and results. Choose approximately one headline result figure from your paper that you will present. Do not attempt to include all the data you collected, or even all the data you described in the paper.
  5. The conclusion. This can often be the “executive summary” from the top of your talk, rephrased as a takeaway message.

This is a reasonable template for full conference talks in systems areas, but it may not apply exactly for position papers, workshop talks, and the like.

Make the introduction slides.

The introduction is the most important part of the talk: it’s the only part of the talk that the majority of your audience will both understand and remember. It will also guide the visual language that you use in the rest of your slides.

Make the “signposting” slides.

Now that you’ve introduced your audience to the topic, you need to give structure to the remainder of your talk.

Fill in the content.

This is the “meat” of the talk, but in some ways it’s the easiest part: you translate the technical content of your paper into visuals and audio.

To give the right amount of detail, make a mental model of the background you expect the conference audience to have. Advisors and students who have been to the conference can help with this. You need to strike the tenuous balance between boring the more experienced people and confusing the outsiders and younger students.

Practice.

Practice relentlessly. In my experience, practice is the primary predictor of talk quality: unpracticed talks are almost always terrible; thoroughly practiced talks usually go well. Every other effect is secondary.

Two kinds of practice are important: