• Table of Contents

Sampling lets us implicitly define a distribution by examples

Surprisingly deep topic in Stats/Prob ML

• You don’t need the pdf/pmf of probability distribution to sample from it
• Sometimes, you don’t even need to materialize likelihood

TODO:

• Langevin MCMC
• Glivenko–Cantelli theorem - Wikipedia

Energy-based

Motivated by having an unnormalized probability that you don’t want to touch directly

Rejection Sampling

MCMC

Misc.

Inverse Transform Sampling