Kolmogorov semigroups are a natural way to combine MCMC methods with Stochastic Gradient Descent.
by Roland Halbig
Kolmogorov semigroups are a natural way of describing time evolution of Itô diffusion
equations. These equations are an important class of so-called Langevin equations fre-
quently used in Markov Chain Monte Carlo (MCMC) simulation and have been shown to
be closely related to the Stochastic Gradient Descent (SGD).
The “code” repository contains everything I needed to create all the plots.
Although most of the code is specific to plotting purposes, the discretizers and samplers might be useful for future work.
They can be found in the code/samplers folder.