Advanced probabilistic modeling: from generative to neuro-symbolic AI

This course will start by offering a rigorous introduction to probabilistic reasoning for modern ML and AI. We will revise classical tractable models such as finite mixtures and extend them to more deep generative models such as variational autoencoders which are infinite (and uncountable) mixture models.
Then, lectures will touch on the problem of performing reasoning and learning in the presence of logical or symbolic constraints, highlighting what are the challenges and the solutions of modern deep learning pipelines.
Lastly, it will discuss recent works on how to automate the construction of efficient and reliable probabilistic models for a given reasoning scenario.