Presentations

Making oral presentations is an essential way to communicate your work. This page gives some general tips on how to prepare and present your talk. For specific suggestions on a number of different kinds of presentations, see our pages on

Talks vs. papers

Overall, a talk should be organized like a paper, and it should answer exactly the same questions (see writing research papers for details). After all, you are presenting the same material! However, the oral and visual medium requires that you present it slighly differently.

In an oral presentation you can present only a fraction of the details in a paper. On the other hand, you can make much better use of emphasis, and communicate your excitement about the work. Why would anyone come to your talk when they can go and read your paper? The answer is that you will communicate much more "between the lines," i.e. express what is really cool about the work, what is difficult, and offer speculations about reasons and consequences of your findings that you are not quite ready to put in writing.

Second, you cannot throw a lot of text at the audience; they are listening to you, and don't have time to read. Whatever text you have on your slides is there mostly to remind you of what you need to talk about. Instead, you should prepare lots of visual material: diagrams, figures, demos etc. Those are expensive on a paper, but they work great in a talk. So, you should arrange your talk around visual material. These points are expanded on below.

Rule number 1: Make sure to communicate your main points

You cannot present all the details in a short talk, and most people will not be able to follow a very detailed talk well. Try to get a few central ideas through as clearly as possible.

When preparing a talk, your job consists of three main parts:

research your topic
distill that information down to a few most important points
present the highlights orally

People rarely skip steps 1 and 3 (although presenters are occasionally known to have slept right through the talk they were supposed to be giving!) Unfortunately, many people do skip step 2, and they end up with not-so-great talks. Step 2 is actually the most important part: it represents the primary service that you are providing to your audience. Otherwise, for e.g. a cs394n topic talk you may as well just hand out photocopies of every paper you came across.

Remember that if you try to say too much, you will end up not communicating much at all. If you lose everyone by rushing through a dizzying array of topics, by having slides crammed with details than no one can take in, or by trying to get across every subtle nuance of a complex idea, then your audience will get very little out of the talk.

Rule number 2: Plan your talk around figures

For all but the most informal presentations, you will want to put together a set of slides that illustrate your points. First, don't start by putting everything remotely relevant into your slides; it will be difficult to cut them down into something manageable later. In almost every case, you will end up with a talk that you cannot finish in a reasonable time, even though everything seems important and relevant.

It is usually better to do the opposite, with a simple procedure:

Collect a set of figures that illustrate what you are trying to say (a table of data, or an equation expressing insight, can be considered a figure as well).
Put each figure on its own slide with a short, meaningful title.
Sort the figures into the approximate order they would appear in a paper.
Go over the slides from the beginning, imagining what you want to say about each one. On each slide write down a few short bullets with the major points that you want the audience to understand about the figure.
If you have more than about four points to make, consider arranging them hierarchically so that you have max four bullets at the highest level--or separating them into groups. Also consider splitting the slide into two and having a separate set of points in each.
Flesh out the set of slides with a few new ones to cover missing points and to make transitions between topics, perhaps tracking down a figure or two to add to illustrate the new points.
Go over the talk again from the beginning until the point of every slide is totally obvious, ideas flow smoothly as the talk progresses, and the main ideas are clear throughout the talk.

This way, you do not waste time coming up with a lot of slides or text that you will never be able to fit into the talk. Note that if you structure the talk around the figures, you do not usually have to face the tough question "How do I find a figure to illustrate this point?". If you don't have a picture for it, the point had better be very important to still bother to make a slide for it.

Rule 3: Use text sparingly

Each slide has a fundamental limit on the amount of information it can contain. One figure per slide, with about 3-4 bullets below or to the side of it, is a good rule of thumb. For the rare slide with no figures, perhaps 4 bullets, each possibly with 2-3 subbullets. You should try to make sure each bullet is at most one line long, and never write text that requires reading an entire paragraph (ok, maybe if you are quoting someone else).

Notice that this is a very severe limitation! A typical 20-minute conference talk has an upper limit of about 20 or so slides. Thus such a talk would only be able to fit about 200 short lines of text at the extreme, i.e. if it were crammed with text and had no figures at all. That amount of text works out to only a couple of pages from a typical book, so you just can't fit everything you want to talk about. (Not to mention that a talk with even that much text would cause everyones' eyes to glaze over instantly.)

To cut things down, you could start with skipping all the information about methodology and implementation, leaving just the motivation, major results, and clear conclusions. This will probably leave a few gaping holes, i.e. methodology and implementation details that absolutely must be mentioned for the talk to make sense. There might also be a few details that are interesting in their own right, i.e. that you really think the audience will be glad to know. If so, just add those very few crucial topics back in. But for the most part, people who need to understand the details will need to study your written work anyway, and there's no need to confuse everyone else during the talk.

Specific preparation tips

Use contrasting colors

It is a good idea to use color on the slides, but make sure that the colors are distinct and have a high contrast with the background. For instance, never present a yellow line or yellow text against a white background -- it will be invisible when projected. Similarly, don't expect people to be able to tell the difference between dark purple and dark blue; everything opaque will look about the same unless it is really big.

Make sure plots are visible

Be sure to use thick lines whenever possible. Most plotting programs default to single-pixel lines, which are easily visible when sitting directly in front of a computer monitor. However, thin lines do not work well in a presentation, because they are difficult to see from a distance. Be particularly careful when plotting multiple curves using different colors of thin lines -- even if the lines are visible, their colors may not be distinguishable from each other unless the lines are thick. (Single-pixel lines are not usually even a good idea for publications, since they tend to disappear when papers are photocopied, reduced for publication, etc.)

Also see Ray Mooney's section on presenting experimental results in his job talk advice. For instance, once people can see it they still need to be told what they are looking at, i.e. what each axis is and what the plot represents.

Use landscape slides well

Most slides are in a wide landscape format. Be sure not to succumb to the tendency to use overly-long lines of text. Using the full page width is appropriate primarily for wide or side-by-side figures rather than for long sentences. Often it is a good idea to fill the left of the page with a figure, then put text bullets on the right. That way the page won't be cut off by peoples' heads, yet the bullets will be short and easy to grasp.

Utilize demos

You should always consider showing a demo in your talk. However, make sure it is well integrated into the talk. That doesn't mean that it has to play inside the slides; switching to a different medium may even make it look more significant. However, you need to rehearse it well, and know exactly what you say and when, and know how long it will take. It should be understandable and illustrate the point well; it is less important to demonstrate cool graphics, your programming prowess, and technical details. The audience will remember a good demo for a long time. It is also a good way to attract them to our web pages for more demos and papers. See e.g. nn group demos, enn tutorial demos, and evolution.ml demos1 and demos2 for a few example demos.

Title slide

It's a good idea to list your collaborators on your title page, so that you don't have to try to remember to mention them at various places during the talk.

Instead of reading your title and authors aloud (which the moderator probably just did anyway), say something interesting about what this work is all about, or why you are doing it, or perhaps how your collaboration came about.

Outline slide

Most of the time, an outline is a waste of time, and puts people right to sleep. However, presenting an outline slide makes sense if your talk has an unusual structure, for example, if you combine two different topics, or several viewpoints that may be hard to follow in a linear talk. Also in a dissertation defense or a jobtalk you often have much more potential topics than you have time for. In that case you can include all of them in an outline, and explain that you will have time to focus only on a subset of them. If you do have an outline, do not just read it aloud; explain why you have the nonstandard structure, give content that's not on the slide, communicate the big picture.

Last slide

If you presented work that was a collaboration with a lot of people, you may want to include a slide in the end (after conclusions) that identifies each of them, includes face images, and a short label indicating their contributions.

Make sure people know that you are now finished with the talk, and invite questions. However, back up from the collaborators to the conclusion slide, so that will be the slide on screen when you start to answer questions. It is a good idea to put a web link to your work on that conclusion slide, so that people have time to take a note of it if they want.

Giving the talk

Use the slides as a reminder of what you have to say; never read directly from the slide.
If you have an in-person audience and the physical setup allows it, use a slide changer and a laser pointer on the projection screen. That way you can step away from behind your laptop and make a better connection with your audience, and share the same focus with them on the projection screen.
However, if all or part of your audience is online, you need to know where you can stand so that they will see you, and use the pointer on the computer so that everyone will see it.
If you run out of time and have to skip slides, really skip them and simply pretend that they are not even there. Do not flip them on for 1/2 a second and say that you won't have time to talk about them!

Be sure to speak clearly and loudly enough for those in the back to hear. Try to make eye contact with the audience, if only so that you will see how they are responding to the talk. You will soon start sensing when you have your audience with you, and when you lose them. Pay attention to such feedback.
Emphasize each big point: make it forcefully, look at the audience, and stop for a while. Let it sink in before continuing. Don't be afraid to utilize one second of silence---it is a great way to emphasize a point. This way even those that are dozing off might notice and remember what you said :-).

Before the talk, make sure that your A/V system works: The slides and demos show up on the project, your slide changer, laser pointer, and the microphone works. At conferences, usually you'll have a chance to do such checks before the session starts (even if you are not the first speaker).
If you tend to get nervous when speaking in public, it often helps if you chat informally with some members of the audience just before the talk, so that the talk itself is just a continuation of the conversation. This helps to build a sense of camaraderie with the audience, and more importantly won't leave you time to let anxiety build up before the crucial moment.
If appropriate, you should schedule a practice talk beforehand in an NN-meeting. Our feedback should help you clarify your slides and approach, and the whole process will hopefully make the actual talk seem less formidable.