Probabilistic Deep Learning

Ahh this is cool.

So like a traditional neural network is purely discriminatory: you feed it data, and it learns to discriminate between two classes (in the case of Classification).

However, with these probabilisitic models, you are trying to learn the underlying distribution of the model. Thus, these models are actually Generative Models, since you know the underlying distribution, you can create new data.

This course seems ok: https://www.coursera.org/learn/probabilistic-deep-learning-with-tensorflow2