This tutorial provides an in-depth explanation of 3 big but related ideas in machine learning - Latent Variable Model, Amortized Inference and Variational Autoencoder.
Most of the time VAE is explained as an Autoencoder where the latent vector has a distribution however that explanation misses the main goals behind its motivation.
First, we see what it means to have an amortized inference and how a certain category of models called Latent Variable Models requires it in order to be efficient when we deal with large datasets.
Then we construct a neural network that addresses the challenges with Latent Variable Models leading to the creation of VAE.
# Recommended videos to watch before this one
KL Divergence
https://www.youtube.com/watch?v=9_eZHt2qJs4
Evidence Lower Bound
https://www.youtube.com/watch?v=IXsA5Rpp25w
#variationalinference
#latentvariable
#variationalautoencoder
