mloss.org OpenGMhttp://mloss.orgUpdates and additions to OpenGMenFri, 12 Nov 2010 17:00:50 -0000OpenGM 1.0 -- Optimization Library for Higher Order Graphical Modelshttp://mloss.org/software/view/283/<html><p>OpenGM is a free C++ template library, a command line tool and a set of MATLAB functions for optimization in higher order graphical models. Graphical models of any order and structure can be built either in C++ or in MATLAB, using simple and intuitive commands. These models can be stored in HDF5 files and optimized using state-of-the-art algorithms and the OpenGM command line optimizer. All library functions can also be called directly from C++ code. OpenGM realizes the Inference Algorithm Interface (IAI), a concept that makes it easy for programmers to use their own algorithms and factor classes with OpenGM. </p> <ul> <li><p>Factor Graph Models (Kschischang et al. 2001) </p> <ul> <li> Graphs of any order and structure, from regular grid graphs to irregular graphical models with higher order factors. </li> <li> Flexible number of labels (different variables can have differently many labels). </li> </ul> </li> <li><p>Optimization Algorithms </p> <ul> <li> Loopy Belief Propagation (Pearl 1988, Yedidia et al. 2000) with message damping (Wainwright 2008), including Min-Sum and Max-Product message passing. </li> <li> Tree-reweighted Belief Propagation (TRBP) (Wainwright et al. 2005) with message damping. </li> <li> A-star branch-and-bound search (Bergtholdt et al. 2009). </li> <li> Sub-Gradient Descent (Dual Decomposition) coming soon! (Kappes et al. 2010).<ul> <li> Automated decomposition of arbitrary factor graphs </li> </ul> </li> <li> Iterated conditional modes (ICM) (Besag 1986). </li> <li> Lazy Flipper (Andres et al. 2010). Binary variables only. </li> <li> Graph Cut (Boycov et al. 2001). Push-Relabel (Goldberg and Tarjan 1986), Edmonds-Karp (Edmonds and Karp 1972), Kolmogorov (Boykov and Kolmogorov 2004]). Binary variables, 2nd order models and submodular functions only. </li> </ul> </li> <li><p>Command Line Optimizer </p> <ul> <li> Built-in protocol mode for runtime and convergence analyses. </li> </ul> </li> <li><p>MATLAB Interface </p> <ul> <li> Build your graphical models conveniently in MATLAB. </li> <li> HDF5 Import/Export </li> </ul> </li> <li><p>High Performance Computing </p> <ul> <li> Optimization of graphical models that consist of 10^7 factors and more. </li> <li> Optimized class templates for binary variables (contributed by Thorben Kroeger). </li> </ul> </li> <li><p>Extendibility </p> <ul> <li> Add and contribute your own optimization algorithms and factor classes. </li> </ul> </li> </ul></html>Bjoern Andres, Joerg H. KappesFri, 12 Nov 2010 17:00:50 -0000http://mloss.org/software/rss/comments/283http://mloss.org/software/view/283/bayesian networksfactor graphsgraphical modelsmarkov random fieldsbelief propagationdiscrete optimizationhigher order cliqueshigher order factorsinference