Drag to orbit · right-drag to pan · scroll or pinch to zoom · tap blocks
Before you start
Why the books have so many equations
An equation is a sentence squeezed down. z = w·x + b means "multiply each input by how much it matters, add them up, then add a nudge." The books aren't writing garbage. They are writing three kinds of thing, and you only need the first kind to build:
Recipe
What to compute. There are about six of these, listed below. Everything in Murphy 1, D2L and most of Murphy 2 is these six, combined, plus one new idea (probability distributions over the knobs themselves).
Bookkeeping
The same recipe written for every index, layer and batch: Σ over i, superscripts for layer ℓ, matrices instead of loops. It looks like new math, but it's the same recipe written for every case at once.
Proof
Why the recipe works or when it fails. "Squared error = assume Gaussian noise" (Murphy §4.2). "A linear autoencoder = PCA" (§20.3.1). You can skip these on the first pass and come back once the simulation has made you ask "but why?"
So the order here is: see it move, run the code, then read the equation as a caption. Every station has an equation you can tap symbol by symbol.