Mixture approximations in science and engineering

These days, there is little need to motivate the use of mathematical models to describe the data we obtain from the world around us. Everywhere we turn, there's a digital twin, a data-driven forecast, or an AI solution that's making smart decisions in real time. However, the underlying data that drives these technologies comes from a place of uncertainty. Classical models disregard this and focus on known phenomena and how they behave. Modern modeling paradigms are different. They use probabilistic methods to account for the uncertainty in data while also taking into account the knowledge of the underlying phenomena.

In this talk, I will introduce the concept of mixture approximations. There are three basic kinds: Gauss, Erlang, and beta. The first is well-known and is often called a Gaussian mixture model (GMM). The others are less common but have equal merit. We will see how to use the approximations to model a data set and quantify uncertainty, and I will show examples from our research at DTU.

Link (mixture approximations): https://tobiasritschel.github.io/mixbox

Link (professional website): https://people.compute.dtu.dk/tobk