The Probabilistic Graphical Model (PGM) Library is a C++ template library for statistical inference in higher order factor graph models. The PGM library implements the Inference Algorithm Interface (IAI), a specification for combining different classes of graphical model factors with different inference algorithms. The current release features factors which depend on an arbitrary finite set of variables each of which has finitely many states. Loopy belief propagation (BP) with synchronous message passing and message damping is realized in an IAI compliant fashion, with the sum-product, max-product, and max-sum variant illustrating the flexibility of the IAI. MATLAB users find code for the construction of factor graph models with explicit functions from inside MATLAB, and MEX functions to call the library's BP algorithms. A simple binary file format is specified for explicit graphical models such that these models can be exchanged with programming environments such as R.
Last update: 08.10.2010, 11:00