Model-Based Machine Learning

Skimming this, it mostly focuses on Bayesian probabalistic models. I think these are underutilized in machine learning tasks, especially since it can be so easy nowadays with tools like PyMC and fast variational methods. If you have a reasonable data-generating model in mind, that information should be incorporated into the prior.

Yesterday: https://news.ycombinator.com/item?id=18201986.