Preparing for Talks: research comps, thesis proposals and defenses

Technicalities

For both research comps and thesis talks, do the following:

post to general (now) that you'll be doing this; include your abstract.
Re-post to general, and send mail to fac and Phd about 3 days before your talk.
Have the front desk send a "wall" 5 minutes before your talk.
Round up the relevant faculty on your committee so that the talk can begin on time.

General guidelines for thesis proposals:

(0) Remember that you have to give in a written document to your committee (length 10-30 pages) and then give a proposal talk after 4 weeks. And your committee usually has 3 people in it (including your advisor). At most one can be from outside the Univ.
(1) It's supposed to be 1 hour with questions. So plan your talk for about 40 minutes.
(2) Make the motivation accessible to everybody, so that they all feel like they want to listen to your talk. Make it really technical if you don't want them to ;-)
(3) Be fair to your related work. You don't have to bash it needlessly to make your work look better, but don't spend too much time on it (about 1-2 minutes is enough).
(4) There is no time for full proofs in your talk. Instead give proof sketches which convey the gist of the idea.
(5) You can convey about 3-4 ideas in the time that you have.

Detailed suggestions (for thesis proposals, but worthwhile for other talks too):

What follows is a not-very-well-organized collection of thoughts about Ph.D. thesis proposals and defenses. It starts with my summary of expectations of the presenter and of the faculty and others in attendance. At the end are some suggestions for how to best give a proposal talk.

Thesis proposals and defenses -- a position statement

I want to clarify something about Ph.D. thesis proposals and defenses and committees. Thesis defenses should be an opportunity for students to present their work to an audience that expects success; the work should be done, the committee should have tacitly approved it already, and the event should be one in which the audience is educated and impressed by what's presented. The student should feel confident and give a good and clear talk. S/he will, with luck, get probing questions from the audience to which s/he has the answers, and perhaps (with even more luck), get interesting suggestions for future work or applications. I believe that if the defense is a pitched battle against an onslaught of difficult questions about the merit of the work, the advisor and or committee has failed -- the defense should never have been scheduled.

BY CONTRAST, thesis proposals are a time for course-corrections. They provide a chance for the faculty to maintain departmental standards, and to make sure that (for instance) if there's work in another field that's closely related to what the student proposes, the student go and learn it and learn from it, so as to not present a final defense in which some chance attendee stands up and says "but this is all just stuff that Milgram did in databases 20 years ago, dressed up as research in AI!"

During a proposal, students should expect to get questions that they cannot answer (if they can answer all the questions, the proposal has come too late in their graduate career). They should expect to be quizzed on what they're saying, so that anyone in the audience who WANTS to follow the presentation closely can do so. They should expect to get suggestions about limiting or expanding the scope the the planned work, about its applications, about its generality, about its originality and relation to prior work. The talk should be aimed at a general CS audience, at least at the start when the motivation is presented. You, as presenter, cannot assume that everyone knows that in graphics, building space- and time-efficient space- partitioning data-structures is important for interaction, rendering, collision-detection, and making fast video games. You have to say so. And if you don't, you should expect your audience to ask you questions until you convince them that it IS important.

In short, if you're planning to get nervous about one of these presentations, either as student or advisor, it should be the proposal rather than the defense.

By the way, I am told that there's a departmental policy (at least "it's been done this way a lot") that says that the dep't will pay travel expenses for an outside committee member to attend one of the two events. My own strong recommendation is that it be the proposal rather than the defense.

Tips for giving a good proposal or defense talk

First of all, let me say that there are as many styles of speaking as there are speakers, but that there are far fewer GOOD styles of speaking. If you are uncertain of your own style, you can often do a B+ kind of talk by mimicking someone else whose style you know is good.

Second, a talk should have a plan. Any talk is an attempt at persuasion; in the case of a thesis proposal, you're trying to persuade the audience that the proposed work is worthwhile, that you can do it, and that when you're done, it'll be substantial and significant. Word to the wise: this persuasion can usually be completed without almost no reference to technical details.

Nonetheless, as scientists we like our technical details. They're what we're good at, what we know ... talking technical is our "comfort zone." Sadly, it's not usually the audience's comfort zone, so you have to compromise. The talk is for the audience, not for you.

If you need a security blanket -- something to help you feel you're comfortable, you're within a pleasant structure -- then you can use the repeated-slide approach, in which, after the first few slides, you present a roadmap to your talk in which there are 5 bullets (intro, related work, proposed work, details, contribution, for example), and one of them is an arrow. As you work through the talk, you show that slide again and again, with the arrow moving down. This helps your audience know when you're done, which matters. On the other hand, if you're confident of your speaking skills, this approach may get in your way -- don't feel obliged to use it.

Use technology to help your talk, but for heaven's sake, KNOW the technology. Don't project slides from the computer if you've used 10-point font. In FrameMaker, I use 18 point font on an 11 by 8.5 inch layout, and then enlarge it to full screen; this turns out to work OK for ordinary text on a slide; I use 24 or 30 point for titles.

Have a conclusion. "Thank you very much" is a sort of weak ending. We all know you're done, but ... Far better is to say something like "And so, in conclusion, we see that frangible widgets are not just a theoretical curiosity, but instead are the pathway to the future of computing." You say this with a rising, then falling pitch to your voice, clearly and looking straight at the audience, having taken a step or two towards them, away from the spot where you've been lecturing all along. This motion to a new locale will tell them very clearly that something different's happening, and the "in conclusion" part will let them know, at then end of this clearly-spoken sentence, that you've finished.

Here are some less-thoroughly spelled out suggestions. If you're puzzled by them, talk to me.

Make eye-contact with your audience.
Roughly speaking, the first third of the talk should be understandable by everyone; the second third by all faculty; and the last third by experts.
Lay out the background as early as possible and never use undefined terms.
Definitions are best given by everyday English and examples (optionally followed by formal definitions). Show people something and point out something interesting about it and say "THIS is what we mean by a frangible widget ... more formally, ..." People are GREAT at generalizing from examples.
Spellcheck your slides! A mis-spelled word says to your audience "I didn't consider this important enough to worry about getting that sort of thing right."
Number your slides by "current page number/total page number."
Define your problem as early as possible.
Remember that slides complement the talk. Don't read from your slides.
Avoid using a sliding piece of paper to disclose more and more of a slide (this is considered annoying to some people). On the other hand, if you're using electronic slides, having the same slide repeated multiple times, each time disclosing another bullet, can work quite well.
Figure out your goal(s) before the talk; then be sure the talk meets them.
Don't be afraid of boring the experts. They're going to be bored anyhow, so by trying to amuse them, you merely lose the REST of the audience.
Give practice talks and listen to the feedback you get.
Don't put too much information on a slide; two or three bullets is often enough; six bullets is too much.
Use a sans-serif font, except for paragraphs (of which you should have very few; an occasional motivating quotation, a formal definition...)
Point at the screen, not the overhead projector, when you explain things.

Compiled by John Hughes (Spike), Brown Grad Rep ? - '02
On behalf of the grads, thank you, Spike.