This project aims to develop novel Message Passing (MP) methods for signal processing applications. Many inference problems of practical interest in signal processing involve multidimensional marginalization procedures, and MP methods are readily employed to alleviate such computational burdens. MP methods harness the statistical dependencies between several variables or nodes. By exchanging carefully-designed messages between these nodes, MP methods can lead to computationally efficient inference algorithms.  

However, choosing the right implementation for the messages is still a challenge, especially when dealing with non-linear and non-Gaussian models. As part of this project, various representations for these messages will be investigated, such as, parametric families of functions or particle-based approximations. Furthermore, a comparison will be conducted of these implementations for multi-object tracking in challenging scenarios.