I’m a member of the Probabilistic Programming Language team, currently working on developing the Bean Machine library for Bayesian analysis. Prior to this, I worked on natural language understanding, especially for building models that can reason in a human-like way. I really like the Bayesian approach because it allows people to easily incorporate domain knowledge as priors. Additionally, the results of Bayesian analysis are generally very interpretable. The combination of neural networks and Bayesian network should allow us to take advantage of the pros from both sides and do more than either of them is able to do by itself.
Probabilistic graphical models, human-like cognitive models, natural language understanding