Gradient descent and neural networks built from rounding errors
suckerpinch · Explainer · 55 min · 2023 · YouTube title: GradIEEEnt half decent: The hidden power of imprecise lines
Why it's here
Cited as a fun, unconventional example of neural networks built from floating point error.
What it covers
Tom 7 (suckerpinch) delivers a long, comedic digression on gradient descent and floating-point imprecision. He shows how rounding behavior can supply the nonlinearity needed to build a neural network.