Probabilistic Programming and Statistics
Webinars
Click on each webinar for its recording and materials. Bayesian statistics is more intuitive to the way we (humans) think about the world and easier to interpret than the traditional frequentist approach. Moreover, it allows for the incorporation of prior information and diverse data, it provides a measure of uncertainty, and it is extremely valuable in difficult situations with little data. The downside is that it is computationally complex and intensive. In addition, the best performing algorithms require burdensome calculations of derivatives. This explains its initial limitations. With the advent of increasingly performant probabilistic programming languages (PPLs), algorithms, compilers, automatic differentiation engines, and computer hardware, Bayesian approaches are now fast growing in popularity in many fields. JAX is a library for Python that makes use of the extremely performant XLA compiler, runs on accelerators (GPUs/TPUs), provides automatic differentiation, just-in-time compilation, batching, and parallelization. In short, it is a perfect tool for Bayesian statistics. Not surprisingly, many PPLs now use it as a backend, and Stan users (and developers!) are turning to it. In this webinar, I will give a brief and very high-level introduction to Bayesian inference, then talk about the various PPLs and samplers which use JAX. Bayesian inference in JAX (2025-Feb-25)