@inproceedings{e7ab9c9cdddd439eb057feff99826830,

title = "What cannot be learned with bethe approximations",

abstract = "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.",

author = "Uri Heinemann and Amir Globerson",

note = "Funding Information: This study was supported by a Studentship award from the British Heart Foundation (FS/96045). We are also indebted to Professor Peter Mobbs for help with the epifluorescence microscopy and William Hinkes for technical assistance with the confocal microscopy.",

year = "2011",

language = "אנגלית",

series = "Proceedings of the 27th Conference on Uncertainty in Artificial Intelligence, UAI 2011",

publisher = "AUAI Press",

pages = "319--326",

booktitle = "Proceedings of the 27th Conference on Uncertainty in Artificial Intelligence, UAI 2011",

}