Statistical relational learning is a branch of machine learning (A.I.) devoted to unify probability theory and logic. I’ll write another post later to explain the motivation and a bit of history of this fascinating branch of study, but here I want to focus on a concrete example, with detailed maths and code.
The approach to statistical relational learning explained here is called Markov logic network (MLN), discovered in 2006 by Richardson and Domingos. Their paper has a nice simple example of MLN applied to the relationship between smoking and cancer. However, it’s a bit hard to follow unless you’re used to read papers on both logic and probabilistic graphical models. In this post, I will mostly follow their smoking/cancer example, but I will try to be much more explicit. I’ll also do a demonstration with Manticore, a small implementation I wrote for playing with statistical relational models.