Estimating the rate-distortion function of real-world data, part 2
Your (beta-)VAE might be secretly estimating the rate-distortion function of your data.

Estimating the rate-distortion function of real-world data, part 1
What is the rate-distortion function, and why we may care about it.

The Ill-defined Problem of Maximum Likelihood Estimation
Why we use maximum likelihood for density estimation, when it breaks down (especially on real-world data), and what can be done about it.

All the Ways to Carve Up the ELBO
My list of fancy decompositions of the aggregate ELBO, focusing on the role of the aggregate KL regularizer, its relation to the aggregate posterior, and the mutual information between the data and the latent variable.

Variational Autoencoder
Training a VAE on MNIST from scratch using Keras.

Projections of Probability Distributions and the Reparameterization Trick
A brief review of M-projection, I-projection, REINFORCE, and the Reparameterization Trick.

Restricted Boltzmann Machine and Contrastive Divergence
RBM training with CD from scratch on MNIST data.

Kernel => RKHS + Feature Map
Given a kernel function, find the Reproducing Kernel Hilbert Space and the feature map it defines.