What Cannot be Learned with Bethe Approximations

Details What Cannot be Learned with Bethe Approximations What Cannot be Learned with Bethe Approximations

2012-02-14

arxiv.org/abs/1202.3731v1

Computer Science

Learning

Scientific paper

We address the problem of learning the parameters in graphical models when inference is intractable. A common strategy in this case is to replace the partition function with its Bethe approximation. We show that there exists a regime of empirical marginals where such Bethe learning will fail. By failure we mean that the empirical marginals cannot be recovered from the approximated maximum likelihood parameters (i.e., moment matching is not achieved). We provide several conditions on empirical marginals that yield outer and inner bounds on the set of Bethe learnable marginals. An interesting implication of our results is that there exists a large class of marginals that cannot be obtained as stable fixed points of belief propagation. Taken together our results provide a novel approach to analyzing learning with Bethe approximations and highlight when it can be expected to work or fail.

Affiliated with

Also associated with

No associations

Rating

What Cannot be Learned with Bethe Approximations does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with What Cannot be Learned with Bethe Approximations, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and What Cannot be Learned with Bethe Approximations will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFWR-SCP-O-90494