D. Blei, A. Kucukelbir, J. McAuliffe. Variational Inference: A Review for Statisticians. 2016.
- Takeaway 1
- Takeaway 2
- Open question or critique
What is this paper showing? What are the contributions? What kind of paper is it: theoretical, applied, something in the middle? Where does it fit into the prior body of work?
- Basics
- Variational inference is an optimization problem (vs sampling in MCMC) using KL divergence
- Pros: VI better on large datasets, faster
- Cons: no guarantees on variance, not exact samples
- Simplest example
- "evidence" is demominator of p(z|x), aka p(x), exponential to calc
- evidence lower bound (ELBO) allows us to get around difficulty calculating p(x)
- mean-field assumes all latent variables are mutually independent
TODO