Musings/Soapbox

Don't make my mistakes.

A collection of writings on miscellaneous topics that will grow over time. Thoughts on anything from emerging research and technology, academic advice, writing tips, and so on.

And what's the deal with "deep learning," anyway? You put in two layers and you call that deep? The only other thing that's two layers and called deep is pizza. Maybe we should call it Chicago-style machine learning instead.

Writing Tips, vol. 1

October 16, 2023

A while back I had started up a Twitter account on academic communication, which was pretty fun! Except that Twitter had become a dumpster fire. I mean, it had always been a dumpster fire, but it became an even bigger and more dumpster-y fire when it was taken over by Elon Musk. I didn't really feel comfortable hosting content there. I'll probably migrate that to BlueSky or some other platform eventually, but I think the tips I had written down were pretty compact and might be useful for some people.

Here were all of my tips, so you don't have to click onto Twitter. Not in any particular order.

When giving a talk, your high-level ideas are usually more important than what you did. Try to change the way your audience thinks about (solving) a problem; don't focus on the details of what you did. The audience can read the paper for that.
It's not enough to deliver information, you also need to be passionate about it. If you aren't excited about your own work, nobody will be.
The talks that I remember best were the ones with a sense of humor.
It's better to have more slides with less information on each than fewer slides filled with content. The more your audience is reading at a time, the less they're paying attention to what you're saying.
I've often seen people use pictures or videos of results for explaining a problem or concept. A simple cartoon is often better, even (especially?) in highly technical work, as it abstracts away confusing details from untrained eyes.
When writing a rebuttal, you don't always have to play defense/counter the negatives. You can also quote and re-emphasize positive remarks by the reviewers. When two reviewers are split on part of the work, you can quote the positive reviewer against the negative one.
With the exception of citing work as direct motivation, the introduction and related work sections of a paper should be distinct. Otherwise, you risk confusing the reader on what is your contribution and what is somebody else's.
Some people like to give their paper a nickname. "<Method Name/Nickname>: This Is What The Method Does" Do a google search on that nickname before you publish it and make sure it's not taken or it will be hard to find your paper.
A good technical talk on a new result should be like a good magic trick - first, you must convince your audience that the problem is impossible*, then, you show the solution.
- *it obviously can't be impossible, and that's the main difference between a magic trick and science/engineering.
In most cases, don't reveal text on a slide until you're ready to talk about it.
In presentations with Q+A, the last slide will be up there for a long time. Make it something you want your audience to remember. It's also a good place to leave contact information.
The hardest thing to revise is a blank page.
- Corollary: If you're stuck, write anything, even the worst version of a section or manuscript.
Engineering is communication.
In a talk or a paper, a lengthy conclusion that does little but recapitulate is unnecessary. Everyone just read you or heard you for a long period of time. Use your last minutes/paragraphs to build upon your presentation, and inspire.
Posters are just advertisements for your research. Most should have three goals: 1. Get the attention of the right people, 2. Convey the high level ideas, and 3. Convince people to read your research. It is not a place to reproduce your paper.
Adding figures can not only help explain methods more compactly and concretely, but help both a) break up text, and b) make certain sections of text more memorable. People will remember a slide or page better if it has a distinctive figure.
Avoid the word "novel," especially in a title. If your research is invention, then what you are doing is novel almost by definition. Use descriptors that are more specific to your work.
Put yourself in the shoes of the world's saltiest reviewer. Start from their biggest complaint with the paper, and work your way to their nitpicks, and preemptively address their complaints with evidence or arguments.
Don't meander in an introduction; people have short attention spans. By the end of the first paragraph or the first few slides of a talk, a reader/listener should know exactly what your work is about.
Be brief. (But make sure to include all of the important details!)

Page updated

Google Sites

Report abuse