You probably heard about the cool things you can do with GANs and VAEs and some of you may even heard about flow-based generative models - but how well do you really understand how they work behind the scenes? What are their inherent strengths and weaknesses?
In this talk we will start with the basics - what are generative models and how each type of them works. Then we will dive into each one of them: the math and probabilistic interpretation, pros and cons compared to others, and see how SOTA papers are handling the limitations of the vanilla method they are based on.