On Improving the Efficiency of the Iterative Proportional Fitting Procedure

Yee Whye Teh, Max Welling
Proceedings of the Ninth International Workshop on Artificial Intelligence and Statistics, PMLR R4:262-269, 2003.

Abstract

Iterative proportional fitting (IPF) on junction trees is an important tool for learning in graphical models. We identify the propagation and IPF updates on the junction tree as fixed point equations of a single constrained entropy maximization problem. This allows a more efficient message updating protocol than the well known effective IPF of Jiroušek and Preučil (1995). When the junction tree has an intractably large maximum clique size we propose to maximize an approximate constrained entropy based on region graphs (Yedidia et al., 2002). To maximize the new objective we propose a "loopy" version of IPF. We show that this yields accurate estimates of the weights of undirected graphical models in a simple experiment.

Cite this Paper


BibTeX
@InProceedings{pmlr-vR4-teh03a, title = {On Improving the Efficiency of the Iterative Proportional Fitting Procedure}, author = {Teh, Yee Whye and Welling, Max}, booktitle = {Proceedings of the Ninth International Workshop on Artificial Intelligence and Statistics}, pages = {262--269}, year = {2003}, editor = {Bishop, Christopher M. and Frey, Brendan J.}, volume = {R4}, series = {Proceedings of Machine Learning Research}, month = {03--06 Jan}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/r4/teh03a/teh03a.pdf}, url = {https://proceedings.mlr.press/r4/teh03a.html}, abstract = {Iterative proportional fitting (IPF) on junction trees is an important tool for learning in graphical models. We identify the propagation and IPF updates on the junction tree as fixed point equations of a single constrained entropy maximization problem. This allows a more efficient message updating protocol than the well known effective IPF of Jiroušek and Preučil (1995). When the junction tree has an intractably large maximum clique size we propose to maximize an approximate constrained entropy based on region graphs (Yedidia et al., 2002). To maximize the new objective we propose a "loopy" version of IPF. We show that this yields accurate estimates of the weights of undirected graphical models in a simple experiment.}, note = {Reissued by PMLR on 01 April 2021.} }
Endnote
%0 Conference Paper %T On Improving the Efficiency of the Iterative Proportional Fitting Procedure %A Yee Whye Teh %A Max Welling %B Proceedings of the Ninth International Workshop on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2003 %E Christopher M. Bishop %E Brendan J. Frey %F pmlr-vR4-teh03a %I PMLR %P 262--269 %U https://proceedings.mlr.press/r4/teh03a.html %V R4 %X Iterative proportional fitting (IPF) on junction trees is an important tool for learning in graphical models. We identify the propagation and IPF updates on the junction tree as fixed point equations of a single constrained entropy maximization problem. This allows a more efficient message updating protocol than the well known effective IPF of Jiroušek and Preučil (1995). When the junction tree has an intractably large maximum clique size we propose to maximize an approximate constrained entropy based on region graphs (Yedidia et al., 2002). To maximize the new objective we propose a "loopy" version of IPF. We show that this yields accurate estimates of the weights of undirected graphical models in a simple experiment. %Z Reissued by PMLR on 01 April 2021.
APA
Teh, Y.W. & Welling, M.. (2003). On Improving the Efficiency of the Iterative Proportional Fitting Procedure. Proceedings of the Ninth International Workshop on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research R4:262-269 Available from https://proceedings.mlr.press/r4/teh03a.html. Reissued by PMLR on 01 April 2021.

Related Material