What a Generative Models?

In this blog, I am going to summary the 5 most common generative models: Auto-Regreesive Model, Variational AutoEncoder, Engery Based Model, Flow Model and Diffusion Model. I will also display the pros and cons between different models, and how we can combine those models to get better performance

What is the Generative Models and Generative AI?

Generative models, as the name indicated, are models that can generative new content. Unlike discriminate models, the generative models are sometime hard to train. But why we need generative models in the first place? We want generative models because:

• Density Estimation:

Density Estimation is the type of task that

Figure 1: Summary of various kinds of deep generative models. (Image Source: Probabilistic Machine Learning)

There are 6 types of generative models, as showed in the Figure 1. In this article, we will go through those 6 different types. Get into the details of each different types and compare those models. How to combine those model to get more complex and useful generative models.

Auto-Regressive Models

Variational Infrence Models

Generative Adversial Networks

Energy-Based Models

Flow-Based Models

Diffusion Models

Combining Different Models

Diffusion Models + VAE

Auto-Regressive + VAE

Auto-Regressive + Flow Models

Flow Model + VAE

Flow Model + GAN

VAE + GAN

Conclusion

In the blog, we has explore go through several generative models. We explore why we need generative models. For different purposes, we can different choice of the models. On the other hand, we can combine different models to get better performance. There are still more room for the generative models.