Lo-Fi Machine Learning
May 23rd 2023
======================

A coworker and I have started working on a project
that gets me back to my scrappy, DIY, low-power cs
interests: lo-fi machine learning.

God I wish I'd come up with that title but,
unfortunately, I didn't. It's an outgrowith of
stuff we'd been talking about for awhile
though. Questions like

* is it possible to do interesting things in
  machine learning that don't require powerful
  hardware?

* can you teach ML without having to teach a
  framework with a bunch of dependencies that will
  be obsolete in six months?

* can ML be separated from dataset maximalism and
  brought to something more pro-social and small
  scale?

And we don't entirely know yet but part of the
hope is that by starting with a stripped down
library that's not built for Big Machine Learning
but is, instead, built around being understandable
and extensible and implements as much from scratch
as possible: everything from markov processes to
baby's first automatic differentiation library

It doesn't have to be super efficient if we're
focusing on small datasets and intentionall janky
uses of ML for artistic purposes.

So I've been brainstorming a lot of demos and
collecting resources the past few days and we're
going to be writing a grant to help support our
work as faculty members at the college basically
all of next month.

We'll see what happens!