Model-Independent Mean Field Theory for Approximate Inference

[Michael Haft]

The following paper on approximate propagation of 
information is available online: 

     Model-Independent Mean Field Theory as a Local Method 
     for Approximate Propagation of Information

	Michael Haft, Reimar Hofmann and Volker Tresp
             SIEMENS AG, Corporate Technology

                     Abstract

We present a systematic approach to mean field theory in a
general probabilistic setting without assuming a particular model 
and avoiding physical notation. The mean field equations derived 
here may serve as a {\em local} and thus very simple method for 
approximate inference in graphical models. In general, there are 
multiple solutions to the mean field equations. We show that improved 
estimates can  be obtained by forming a weighted mixture of the 
multiple mean field solutions. We derive simple approximate expressions 
for the mixture weights, which can also be obtained by means of only 
{\em local} computations. The benefits of taking into account multiple 
solutions are demonstrated by using mean field theory for inference in 
a small `noisy-or network'.


The paper is available online from: 
  http://www7.informatik.tu-muenchen.de/~hofmannr/mf_abstr.html
Comments are welcome. A modified version of this paper is submitted 
for publication. 

___________________________________________________________________________
Michael Haft      
ZT IK 4              
Siemens AG, CR & D   Email: Michael.Haft at mchp.siemens.de
81730 Muenchen       Tel:   +49/89/636-47953
Germany              Fax:   +49/89/636-49767