• Code Generation
• Probabilistic Graphical Models
• Factor Graph
• Multiprocessor

The goal of this thesis is to develop a framework that is able to efficiently produce inference on a given probabilistic graphical model. This framework should be able to offer high performance on multiprocessor target solution in standard and embedded systems.

After a brief introduction where probabilistic graphical models based on Factor Graphs are introduced, the Belief Propagation algorithm is presented.
Later, the basic design concepts on top of which the framework has been developed are explained, presenting the different multiprocessor approaches spanning from dynamic scheduling with OpenMP to static scheduling.

Finally, a possible implementation is illustrated, evaluating the performances on different architectures devoting particular attention to the target multicore embedded platform.