A c++ library implementing Variational Message Passing 
  - as developed by John Winn (http://johnwinn.org/Research/VMP.html)

Inference on parameters that are can be modelled by the exponential family 
(Gaussian's, Rectified Gaussians, Gamma's, Dirichlet) 
are easily set up and solved by means of Variational Ensemble Learning 
via message passing.

Independant Component Analysis
  - as developed by James Miskin (http://www.inference.phy.cam.ac.uk/jwm1003/) -
provides an interesting example.