Making a map

Deciding what to learn

After spending a decade-long career semi-arbitrarily picking up and learning things — almost, but not quite, a random walk — I like to imagine I can do a little bit better over the next ten years. Writing it out as an experimental newsletter is a way to track progress, maintain a little bit of discipline and a way to stay on brand.

Photo by Marjan Blan | @marjanblan on Unsplash

With experience I’ve realized that

  • one lifetime is nowhere near enough to learn all that I would like to

  • I learn by building; watching videos or taking notes doesn’t stick

  • everything flows much more smoothly if I’m having fun

Of course, it’s also somewhat valuable to have a reason for learning

  • going deep on-demand as I work on a project: this helps set realistic scopes

  • going wide and staying current, just to build an index of what’s possible

  • satisfying my curiosity, and simply following anything interesting

With the core framework laid out; taking stock of where I’m at today:

  • I’m building debugging related tools for ml engineers tackling large problems

  • with a somewhat mechanical understanding of how models are trained

  • and a somewhat deeper understanding of software / systems

  • somewhat topically, anything that would help me model Covid

  • … and to wrap things up, I’ve been fascinated by Monte Carlo methods recently

Where I’d like to be

  • understand ML models end to end: Math, Software & Hardware

  • up to date with the current state of the industry — with habits of reading papers

and the way to confirm that I’ve learned any of this is through tangible projects:

  • explained derivations, exercises for Math;

  • working projects for Software & Hardware;

  • with a lot of introspection and tools around the space.