↓ Skip to main content

12. Miscellaneous Advice

Last updated:

A running grab bag of practical advice that doesn’t neatly fit anywhere else.

Posters
#

A poster is a visual medium, so make the figures big and keep text short. Plan for 3 to 4 iterations with me before printing, and send drafts at least a few days ahead so there’s actually time to change things.

A few things I almost always check before you print:

  • Put your name near the top, and leave breathing room around the edges so the title isn’t hugging the border.
  • Keep logos out of the tight corners and away from the title.
  • Keep font sizes consistent (nothing random should be bigger than your title), and bold key statistics in larger type so people walking by can scan them.
  • Group your visuals by idea (for example: inputs, labels, their interaction, and comparison to other datasets) instead of dropping in every plot you made.
  • Match plot backgrounds to the section background, and check at full resolution that nothing is blurry.
  • Standard sizes are usually 48x36 or 36x48 inches, though always check the conference board dimensions first.

Resources and examples of aesthetically pleasing posters:

For printing around U of T, UTPoster on campus is the standard spot, and Staples is another option for same-day printing.

Talks and slides
#

Putting real effort into a talk pays off more than people expect: a sharper talk leads to a better RCM presentation and a cleaner paper story.

  • Aim for at most ~15 slides, which keeps a steady pace of roughly one slide per minute.
  • Write slide titles as full declarative sentences, and use the subtitle for the core takeaway (what someone should learn if they only caught that slide).
  • Slides are a visual medium too, so lean on graphics. Corollary: if you put text on a slide, expect people to read it instead of listening to you. Stick to short phrases, single-line facts, or key-value pairs.
  • Include compact citations right on the slide (Title, ‘YY, doi).
  • Look at your audience. Part of a talk is performance, which means moving around a bit, varying the tone of your voice, and making it engaging.

Two short reads I recommend:

Personal websites and headshots
#

Build a personal website early, even a very simple one. It gives you visibility when people look you up after a talk, poster, or paper. Free options include GitHub Pages (Lexi Jones’s walkthrough is an easy start), Google Sites, or Hugo with Blowfish (what our lab website uses).

Whenever the department, an institute, or a conference has a headshot photographer, take the offer. One of my best photos came from accidentally coming into the office on a day they had a photographer, and I still use it all the time. Having a good photo also helps you visualize yourself in future roles.

Local conferences and volunteering
#

Submit talks and posters to local workshops and conferences (U of T events, Acceleration Consortium, Vector Institute, Canadian chemistry and engineering meetings). They’re low-bar to enter, great practice, and a good way to support the Canadian research ecosystem. If registration is tight or closed, check for student volunteer slots: volunteering usually gets you free registration and makes it easy to meet people.

Practical ML references
#

  • Google Research’s Deep Learning Tuning Playbook: practical, opinionated defaults for choosing batch sizes, sweeping learning rates, and debugging training runs.