Appendix B — Math Refresher

B.1 Probability for diffusion

B.2 The linear algebra of attention

B.3 Convolution as matrix multiplication

# Math Refresher {#sec-math}

## Probability for diffusion

<!-- TODO: Gaussians and their closure properties; KL divergence; the reparameterisation trick; just enough of the ELBO -->

## The linear algebra of attention

<!-- TODO: dot products as similarity; softmax; why scale by sqrt(d); matrix view of multi-head attention -->

## Convolution as matrix multiplication

<!-- TODO: Toeplitz view; transposed convolution without mystery -->