Approximating marginals using discrete energy minimization Workshop Paper

Author(s): Korc, Filip; Kolmogorov, Vladimir; Lampert, Christoph
Title: Approximating marginals using discrete energy minimization
Alternate Title: ICML Workshop: Inferning: Interactions between Inference and Learning
Title Series: Workshop at ICML
Affiliation IST Austria
Abstract: 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.
Date Range: June 30, 2012
Conference Location: Edinburgh, UK
Publisher: ICML  
Date Published: 2012-06-30
Copyright Statement: Copyright retained by authors. Free access.
Open access: yes (repository)
IST Austria Authors
  1. Christoph Lampert
    87 Lampert
  2. Filip Korc
    1 Korc
Related IST Austria Work