How to give a talk Tea – June 18 , 2019

How to give a talk Tea – June 18, 2019

Why is this important? There are at least three things you need to make it in science: You have to be smart. You have to work hard. You have to be able to communicate well.

Why is this important? There are at least three things you need to make it in science: You have to be smart. You have to work hard. You have to be able to communicate well. Consider two classic examples: Stephen Grossberg Henry Markram

Why is it hard to communicate well? • Talks are an absolutely horrible way to communicate. • People can remember a very small number of new pieces of information. • People have a hard time listening and thinking. • Your audience almost always has a broad range of levels.

p(level) level

p(level) level It’s not easy to design a talk that everybody gets something out of.

Gatsby! p(level) level It’s not easy to design a talk that everybody gets something out of. We’re going to handle this with just a few, easy to remember ideas.

What people want to get out of a talk: People do not want to just absorb information. What they want is to construct a story. That means you have to build, incrementally, on what they already know. You have to make conscious decisions about the expertise of your audience!

I generally assume that people can: understand simple geometrical objects (lines, surfaces, ..) know most of the standard probability distributions can make sense of simple ODEs can make sense of integrals have an intuitive feeling for optimization, including Lagrange multipliers know about standard inference algorithms (MCMC, variational, EM, …) know a little bit of neuroscience and probably a few others I forgot

There’s a tendency to assume that the audience knows what you know. That’s almost never true! Fortunately, almost anything can be explained in terms of a few simple ideas. To your grandmother. If your talk requires lots of complicated mathematical machinery, odds are you have not thought about it hard enough!!!!!

That’s really all you need to know. But just to hammer home this point, I have a few simple rules. OK, lots of rules, but only the first couple are really important.

Rule #1: At the beginning of your talk, tell your audience what big picture problem you’re working on. And why it’s interesting. And why it’s new. Every single person in the audience should understand this. If you succeed, people will be much more motivated to listen to your talk! And they will know what to pay attention to!

Rule #2: Your audience should always know why you're telling them whatever it is you’re telling them. They should never, ever, ever have to guess. That’s because people can remember a very small number of new pieces of information. They’re going to make an effort to remember something only if they think it’s important.

Rule #2a: Do not give extra information. No matter how cool or interesting it seems to you, it will make the talk harder for the audience. Rule #2b: Do not just give a list of facts. Your audience will eventually forget why any of them matter. Rule #2b.1: When giving a journal club, don’t just go through the figures. Tell the story the way the authors should have told it! After about two facts.

Rule #3 (Physics Today, 1991): Use equations sparingly. Use standard notation. p(level) level

Rule #3 (Physics Today, 1991): Use equations sparingly. Use standard notation. Do not write RKHS unless you’re going to use it. translation: never write RKHS.

Rule #3 (Physics Today, 1991): Use equations sparingly. Use standard notation. Use consistent notation. Distinguish between scalars, vectors, matrices, and other objects! A good rule of thumb: don’t use matlab notation. And never, ever, ever change notation. Explain what people should pay attention to!

Suppose you have a slide like this: p(sj= 1) 1- T log = icjwji (Ri - 1) - log p(sj= 0)  dcj log p(cj) τ T = isjwji (Ri - 1) + + ξj dt cj You have exactly two choices: 1. spend enough time on the slide so that people totally understand it. 2. tell the audience which parts they need to pay attention to. however, pieces they shouldn’t pay attention to shouldn’t be on the slide!

Suppose you have a slide like this: p(sj= 1) 1- T log = icjwji (Ri - 1) - log p(sj= 0)  dcj log p(cj) τ T = isjwji (Ri - 1) + + ξj dt cj So you really have exactly one choice: 1. spend enough time on the slide so that people totally understand it. Whenever you use an equation, write down only what’s necessary, and fully explain it.

Rule #4: make it easy for your audience. - Your job is to distill the information, and present only the essence. - Give examples. - Don’t present the most general case. Even if your theory applies in arbitrary dimensions, present the 1-D case first. Or maybe only the 1-D case.

Rule #4: make it easy for your audience. - Your job is to distill the information, and present only the essence. - Give examples. - Don’t present the most general case. - Provide as much intuition as possible. - There’s a clarity/accuracy tradeoff. Always choose clarity. - There’s a clarity/compact notation tradeoff. Always choose clarity.

Small rules (bit-rate related): 5. People like to be told things they already know. Reduce initials/acronyms to the bare minimum (because they’re hard to remember). 7. Don’t talk for a long time without referring to your slides.

Small rules (general): 8. If there’s a particularly hard point in your talk, practice alternative explanations. 9. Practice your whole talk! More than once! 10. Get feedback (before and, especially, after) your talk.

Rule #11 (Physics Today 1991): Going over your allotted time should be a capital crime!

Rule #12: All rules can be ignored, but think hard before doing so.

More rules: 13. Make 1 point/talk (and paper!!). 14. Watch other talks for presentation style: take the things you like, avoid the things you don't. ask yourself what they could do better. 15. Laser pointer: keep it steady!!! 16. Listen very carefully to questions; give a short answer. As an aside: before asking a question, ask yourself: is this something only I want to know, or does a reasonably large fraction of the audience also care?

There was lots of information here, but it all boils down to two things: • People do not want to just absorb information. • What they want is to construct a story. • It’s your job to distill your (single!) main point down to its essence, so that your audience can extract it. before after p(level) level

Way too many rules • Tell your audience the big picture problem. • It should be obvious why everything in your talk relates to the big picture problem. • Use equations sparingly. • Make it easy for your audience (provide examples and intuition). • People like to be told things they already know. • Reduce initials/acronyms to the bare minimum. • Don’t talk for a long time without referring to your slides. • If there’s a particularly hard point in your talk, practice alternative explanations. • Practice your whole talk! More than once! • Get feedback (before and, especially, after) your talk. • Going over your allotted time should be a capital crime! • All rules can be ignored, but think hard before doing so. • One point/talk.

How to give a talk Tea – June 18 , 2019