ML Paper Challenge Day 38, 39 — Auto-Encoding Variational Bayes

Papers with Code - Auto-Encoding Variational Bayes

Day 38–39: 2020.05.19–20
Paper: Auto-Encoding Variational Bayes
Category: Model/Unsupervised Learning

SGVB (Stochastic Gradient Variational Bayes) estimator

• efficient approximate posterior inference in almost any model with continuous latent variables and/or parameters
• straightforward to optimize using standard stochastic gradient ascent techniques.

Auto-Encoding VB (AEVB) algorithm -> variational auto-encoder

• make inference and learning especially efficient by using the SGVB estimator to optimize a recognition model
• -> perform very efficient approximate posterior inference using simple ancestral sampling
• -> efficiently learn the model parameters, without the need of expensive iterative inference…