Factor graphs and message passing algorithms

Complex modern day systems are often characterized by the presence of many interacting variables that govern the dynamics of the system. Statistical inference in such systems requires efficient algorithms that offer ease of implementation while delivering the prespecified performance guarantees. In developing an algorithm for a sophisticated system, accurate and representative modeling of the underlying system is often the first step. The use of graphical models to explain the working of complex systems has gained a lot of attention in recent years. Stochastic models are often represented by a Bayesian network or a Markov random field. The graphical representation not only provides a better understanding of the system model but also offers numerous exciting opportunities to develop new and improved algorithms. Factor graphs belong to the class of graphical models that serve to explain the dependencies between several interacting variables. They can be used to model a wide variety of systems and are increasingly applied in statistical learning, signal processing, and artificial intelligence.