# Backprop

Backpropagation: how a neural net's loss gradient flows backward through every parameter. The single most important algorithm to *write yourself* before relying on a framework.

## What's here

- [viz/](./viz/) — interactive HTML visualizations.

## Status

[ ] Not started.

## Prerequisites

[`prereqs/calculus/`](../../prereqs/calculus/) — chain rule, partial derivatives.

## Resources

- Karpathy's `micrograd` — best entry point.
- 3Blue1Brown — backprop episodes.