Codes and Expansions (CodEx) Seminar

Keenan Eikenberry (Arizona State University):
Markov Kernels Valued in Wasserstein Spaces

I'll introduce Markov kernels valued in Wasserstein spaces as general objects for studying distribution approximation problems relevant to statistical inference, generative modeling, and probabilistic programming. After outlining an original theory of integration for these objects, I'll define (approximate) Bayesian inverse maps in this setting and discuss ongoing work on convergence properties for a generalized Bayes's Law.