Tips for effective data science talks
In grad school I learned to practice my presentations before giving them. Teaching classes further reinforced the importance of communication and it’s served me well in industry.
Recently I’ve had to give advice to interns for company-wide presentations of their projects. And I realized that I didn’t have a guide to point them to… so now I’m making it.
This isn’t meant to be gospel, they’re just the tips I’ve learned by doing informal testing over the last 15 years or so.
Let’s start with the most important part: you have to put in hard work. But it’ll get easier with practice.
Think of it like writing software: Have you ever written perfect software before testing it on any users? Why would you expect to deliver a great talk without testing it on people? The analogy to iterative development goes deeper: you can try to be smart about anticipating your audience or you can just go get a small random sample audience and try it out.
Consider their feedback honestly without feeling personally criticized. About half of feedback shouldn’t be taken at face value. When someone suggests a solution to you, first identify the problem they’re addressing. There are often easier ways to solve the same problem.
I suggest fielding only clarification questions during the talk and holding all discussion questions until the end. And it’s good to say this upfront: If someone’s lost on slide 2 and doesn’t feel like they’re allowed to speak up, the lack of understanding will often compound and they’ll struggle to understand the next few slides (and sometimes even the whole talk!) You want to allow questions. But you don’t want to enter a 10 minute discussion with 1% of your audience. Have the discussion afterwards.
Managing your technology
Buy a clicker to advance the slides without distraction. At a conference this is normal but it’s even helpful at company talks. You can focus more on engaging the audience and less on overhead.
If you’re on a Mac, set do not disturb mode. The worst is getting a Slack/Hangouts notification mid-talk. Similarly, close Google Calendar or disable interruptive notifications. Or turn off Wifi entirely.
While we’re on the subject of technology, get there early and double-check the projector/TV. If you fumble with the connectors and setup, the audience may start talking amongst themselves and it can be difficult to regain their attention.
Speaking
If you have a microphone, it’s better to speak louder and keep the microphone further from your mouth. That way it doesn’t amplify the sound of your breathing. And it prevents many of the sharp sounds that cheap microphones pick up. You’ll also seem more confident.
In big rooms, quickly confirm that the back of the audience can hear you. It can seem silly but it’s worth 10 seconds.
Respect the audience
Unfamiliar terminology will often lose your audience. Your audience isn’t stupid but they might not know what a Jacobian is, or were taught a different word for it, or aren’t used to English. Andrew Ng’s machine learning course is an example of good technique; he often spends a few extra seconds to make sure that terms are defined and that people of all different backgrounds have an equal chance.
Describing your problem with concrete examples lessens the complexity of terminology and abstract concepts. In many cases, relying on an example for explanation will make the difference between retaining and losing someone.
Maximize the chance that everyone in the room can read your slides. Increase font sizes from defaults.
Some people have poor vision or no vision. So when you’re showing a graph, describe what you’re seeing in words.
Don’t use difficult words when simple words will do (some caveats apply)
Holding the audience’s attention
It’s important to engage your audience. People mirror one another; if you show interest in them, they will often show interest in you. Try to make eye contact with most of the audience at least once.
Animate text so that one bullet at a time appears. For a complex topic this helps your audience focus on the most important thing right now, which is always the last bullet.
Slides aren’t meant to deliver themselves; the text is just there to provide backup.
Prioritize
Copy from the best but give credit. This is something you learn from making lecture slides when you run out of prep time. If you’re trying to explain the difference between supervised and unsupervised learning for example, there’s probably an excellent diagram online.
A common mistake in computer science is making the content overly technical (even at conferences). The purpose of a conference talk is to get people to read your paper. At the end of a short talk, the audience should all know whether the paper is relevant to them or not. At an industry talk, you might instead be explaining new technology to your peers from other fields. You should aggressively prioritize the material you think 90% of the audience needs to know and never sacrifice the 90% just to show off your technical prowess.
On memorization and practice
I don’t memorize my presentations word for word. But I do memorize the feeling of the talk. For me, practicing is about learning how to flow from one slide to another. Then when I’m giving a talk on stage, I naturally follow the flow that feels right. This leads to talks that are controlled but without the stiffness of memorization.
Generally I practice once by going through one slide at a time and pretending to speak. I move my mouth without making sound. This helps me estimate the timing more accurately.
Then I practice before peers. In particular I try to include some that aren’t familiar with the topic and some that are. The feedback from less technical people is usually the most important. You can even use this for networking and most people will be happy to help.
For important talks, on the day of the talk I practice to myself again. If I have a hotel room I speak aloud.
That’s my method but you may find that memorization works for you. Or you might copy Andrew Ng and record yourself.
Ending thoughts
A sign of a bad presentation is that the audience doesn’t ask questions.
A sign of a good presentation is that they ask you questions.
Signs of a great presentation is that they ask you intelligent questions and debate with one another. You’ve taught them something valuable and their curiosity will push you to learn even more.