Variational Message Passing RRS feed

  • Question

  • Hi All,

    I was wondering if you tell me whether the VMP inference procedure is working based on your own random variable definitions and data structures or it has a different kind of implementation far from the probabilistic programming logic. I mean if the source code is similar to the inference functions at R or MATLAB, or it is implemented based on the probabilistic variables and data that at first you created here. 


    Monday, November 2, 2015 9:15 PM

All replies

  • Hi Ali,

    Variational Message Passing (VMP) is a deterministic approximate inference method. Our User Guide talks about it here. You can run it with Infer.NET in a factor graph that you define by connecting random variables and factors using our modelling API. An example of VMP applied to a GMM can be found here. A nice lecture on our inference algorithms (including VMP) can be seen here. Unfortunately, I'm not familiar with any similar procedures in MATLAB or R to make a comparison.


    Tuesday, November 3, 2015 2:39 PM