Backward particle message passing
Henk Wymeersch, Naga V. Irukulapati, Isaac Sackey, Pontus Johannisson, Erik Agrell · 2015
Particle methods are an established way to represent messages and perform message passing in factor graphs. Despite their common use, there are several cases for which messages are hard to compute, even in linear models. Building on results from Gaussian message passing, we demonstrate how backward particle-based messages can be computed and describe a practical application in the context of fiber-optical communications.