) (2)

Here we provide details of the variational inference method for the mixPLDS model. To this end, we first discuss variational inference for the case of a single mixture component M = 1, a model that is equivalent to the Poisson linear dynamical system (PLDS) model defined in Macke et al. (2011). 1 Variational inference for Poisson linear dynamical system 1.1 Notation We first introduce the “vectorized ” notation for the PLDS model. The PLDS is equivalent to the mixPLDS model for M = 1. We therefore drop the group index m when focussing on the PLDS. x:= x1... xT , y:= y1... yT , b:= b..