This toolbox implements the EP algorithms described in "A family of algorithms for approximate Bayesian inference" and "Expectation Propagation for approximate Bayesian inference".
This toolbox requires the Lightspeed toolbox.
See Contents.m for a list of functions. This toolbox uses the object-oriented features of Matlab. You first create a 'task' structure, then create a 'bpm_ep' object. This object has methods like 'train' and 'classify'. See the 'test' scripts for examples.
Tom Minka