ELBO — What & Why | Yunfan’s Blog
ELBO (evidence lower bound) is a key concept in Variational Bayesian Methods. It transforms inference problems, which are always intractable, into optimization problems that can be solved with, for example, gradient-based methods.
ELBO (evidence lower bound) is a key concept in Variational Bayesian Methods. It transforms inference problems, which are always intractable, into optimization problems that can be solved with, for example, gradient-based methods. Updates • May 26, 2021 I rewrote the entire story, added more figures, but left derivations unchanged. Wish that the second version could better help you :). • April 16, 2021 An extensional derivation for the case of temporal sequences has been updated here. Introduction In this post, I’ll introduce an important concept in Variational Bayesian (VB) Methods —…
saved by
related reading
- Variational Inference - Explainedyoutube.com
- Coordinate Ascent Mean-field Variational Inference (Univariate Gaussian Example)suzyahyah.github.io
- Eric Jang: A Beginner's Guide to Variational Methods: Mean-Field Approximationblog.evjang.com
- Mediumandacdemir-94699.medium.com
- Bayesian Neural Networkscs.toronto.edu
- [1906.02691] An Introduction to Variational Autoencodersarxiv.org
- Tutorial - What is a variational autoencoder? – Jaan Lı 李jaan.io
- 1312.6114.pdfarxiv.org
- RescaledKL.tmdanilorezende.com
- Six (and a half) intuitions for KL divergence — LessWronglesswrong.com
- Approximating KL Divergencejoschu.net
- The Reparameterization Trickgregorygundersen.com