Robert Polzin, Freie Universität Berlin

**"What is... DBMR?"**

In this talk, a recent method of Susanne Gerber and CRC 1114 Mercator fellow Illia Horenko, DBMR, is discussed. The method constructs a directly low-rank transfer operator, reducing numerical effort and error due to finite data. Given two categorical random variables with respective ranges, the aim is to find a stochastic matrix of conditional probabilities between the discrete states of both processes. The usual maximum-likelihood estimate of the matrix requires a large amount of pair observations. Gerber and Horenko suggest an efficient and scalable estimation of the transfer operator by introducing intermediate latent states.