We consider the problem of inference in a graphical model with binary variables. While in theory it is arguably preferable to compute marginal probabilities, in practice researchers often use MAP inference due to the availability of efficient discrete optimization algorithms. We bridge the gap between the two approaches by introducing the Discrete Marginals technique in which approximate marginals are obtained by minimizing an objective function with unary and pairwise terms over a discretized domain. This allows the use of techniques originally developed for MAP-MRF inference and learning. We explore two ways to set up the objective function - by discretizing the Bethe free energy and by learning it from training data. Experimental results show that for certain types of graphs a learned function can outperform the Bethe approximation. We also establish a link between the Bethe free energy and submodular functions.
ICML: International Conference on Machine Learning
2012-06-26 – 2012-07-01
Korc F, Kolmogorov V, Lampert C. Approximating marginals using discrete energy minimization. In: ICML; 2012.
Korc, F., Kolmogorov, V., & Lampert, C. (2012). Approximating marginals using discrete energy minimization. Presented at the ICML: International Conference on Machine Learning, Edinburgh, Scotland: ICML.
Korc, Filip, Vladimir Kolmogorov, and Christoph Lampert. “Approximating Marginals Using Discrete Energy Minimization.” ICML, 2012.
F. Korc, V. Kolmogorov, and C. Lampert, “Approximating marginals using discrete energy minimization,” presented at the ICML: International Conference on Machine Learning, Edinburgh, Scotland, 2012.
Korc F, Kolmogorov V, Lampert C. 2012. Approximating marginals using discrete energy minimization. ICML: International Conference on Machine Learning, Inferning 2012,
Korc, Filip, et al. Approximating Marginals Using Discrete Energy Minimization. ICML, 2012.
IST-2016-565-v1+1_DM-inferning2012.pdf 305.84 KB
Material in IST: