Thursday, June 25th, 2020:
Discussion Topic: MCMC and all that: how to do parameter inference and model comparison when things get messy
Lead: Aaron Vincent
Overview: Hypothesis testing is one of the most important parts of scientific research. This is typically expressed as model comparison (is supersymmetry a better model than hyperdupersymmetry?) or parameter inference (what is the most likely value of the sneutralinissimo mass and coupling in my hyperdupersymmetry model?) One of the most challenging aspects in either context is that of sampling. If the parameter space and/or amount of data is large, incorrect sampling can yield spectacularly wrong results, while inefficient sampling can easily keep your CPU busy well past the point when your funding runs out. To address these issues, I’ll introduce the Bayesian framework behind efficient sampling techniques, and describe two common methods: MCMC and nested sampling (Frequentists: don’t panic, you can use these tools too). Time allowing, we’ll end with some kind of tutorial/hands on exercise using emcee (https://emcee.readthedocs.io/en/stable/) and perhaps even Multinest (https://johannesbuchner.github.io/PyMultiNest/pymultinest.html) if we get the time.
Prerequisite: An installation of python and anaconda.