Variational Inference in Generative Models

Variational Inference (VI) first starts as a handy tool in Bayesian inference to approximate intractable posterior. Now, its usage goes beyond Bayesian inference, and we can see VI everywhere in classic learning, and deep learning-anywhere needs an approximation. After this lecture, we should:

• Understand the fundamentals of VI (why “variational” here?)
• See how models use VI (from Topic to Diffusion)
• Use VI for your problem

Part I: Non-deep VI

1. Introduction

2. General_Formulation

3. Classical VI

4. Topic Model

Part II: Deep VI

TBU