
Abstract: This session will go through an overview of the Bayesian workflow for exploring and learning from data while, crucially, retaining and propagating uncertainty. We will see a simple case illustrated and then split out into groups to get our hands dirty with a more complicated hierarchical regression example (radon measurements in Minnesota). There will be a discussion of the ways in which inference algorithms can fail, diagnostics, and procedures for quantifying algorithmic unfaithfulness and for massaging our model's geometry into something more amenable to inference. PyStan or RStan should be installed and usable before the session begins.
Bio: Sean comes from an industry background where he last worked with Watson creator Dave Ferrucci at the Collaborative Intelligent Systems Lab at a New York area hedge fund. He is now one of the core developers of Stan (http://mc-stan.org), a probabilistic programming language that scientists, researchers, and even economists can use to do statistical inference. Interests include programming language design, compiler optimization, epistemology, bayesian data analysis, and helping others do better science.