Bayesian data science

Two arguments for not using Bayesian statistics in a data science context are:

1. Having to wait for several hours of MCMC simulation every time you fix a bug in your model.

2. Worse: having to spend several months implementing finicky MCMC algorithms.

I will talk about the work of my collaborators, students, and myself on trying to make this suffering less severe, using ideas from the disparate fields of statistical mechanics and software engineering.

**Warning:** This will mostly be an unconventional talk: a large chunk will be devoted to introducing "blang", an experimental probabilistic programming language for Bayesian data science we are working on.  **Please bring your laptop with Chrome installed.**