Randomized Optimum Models for Structured Prediction

One approach to modeling structured discrete data is to describe the probability of states via an energy function and Gibbs distribution. A recurring difficulty in these models is the computation of the partition function, which may require an intractable sum. However, in many such models, the mode can be found efficiently even when the partition function is unavailable. Recent work on Perturb-and-MAP (PM) models (Papandreou and Yuille, 2011) has exploited this discrepancy to approximate the Gibbs distribution for Markov random fields (MRFs). Here, we explore a broader class of models, called Randomized Optimum models (RandOMs), which include PM as a special case. This new class of models encompasses not only MRFs, but also other models that have intractable partition functions yet permit efficient mode-finding, such as those based on bipartite matchings, shortest paths, or connected components in a graph. We develop likelihood-based learning algorithms for RandOMs, which, empirical results indicate, can produce better models than PM.

[1]  J. Munkres ALGORITHMS FOR THE ASSIGNMENT AND TRANSIORTATION tROBLEMS* , 1957 .

[2]  C. Smith A Characterization of Star-Shaped Sets , 1968 .

[3]  Leslie G. Valiant,et al.  The Complexity of Computing the Permanent , 1979, Theor. Comput. Sci..

[4]  William H. Cunningham,et al.  Minimum cuts, modular functions, and matroid polyhedra , 1985, Networks.

[5]  G. C. Wei,et al.  A Monte Carlo Implementation of the EM Algorithm and the Poor Man's Data Augmentation Algorithms , 1990 .

[6]  Alejandro F. Frangi,et al.  Muliscale Vessel Enhancement Filtering , 1998, MICCAI.

[7]  D. Dittmar Slice Sampling , 2000 .

[8]  Vladimir Kolmogorov,et al.  What energy functions can be minimized via graph cuts? , 2002, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[9]  Vladimir Kolmogorov,et al.  "GrabCut": interactive foreground extraction using iterated graph cuts , 2004, ACM Trans. Graph..

[10]  Andrew Blake,et al.  "GrabCut": interactive foreground extraction using iterated graph cuts , 2004, ACM Trans. Graph..

[11]  Ben Taskar,et al.  A Discriminative Matching Approach to Word Alignment , 2005, HLT.

[12]  Thomas Hofmann,et al.  Large Margin Methods for Structured and Interdependent Output Variables , 2005, J. Mach. Learn. Res..

[13]  J. Andrew Bagnell,et al.  Maximum margin planning , 2006, ICML.

[14]  Fu Jie Huang,et al.  A Tutorial on Energy-Based Learning , 2006 .

[15]  Zoubin Ghahramani,et al.  MCMC for Doubly-intractable Distributions , 2006, UAI.

[16]  M. Bernardine Dias,et al.  The Dynamic Hungarian Algorithm for the Assignment Problem with Changing Costs , 2007 .

[17]  Ryan P. Adams,et al.  The Gaussian Process Density Sampler , 2008, NIPS.

[18]  Vladimir Kolmogorov,et al.  Feature Correspondence Via Graph Matching: Models and Global Optimization , 2008, ECCV.

[19]  Michael I. Jordan,et al.  Graphical Models, Exponential Families, and Variational Inference , 2008, Found. Trends Mach. Learn..

[20]  Derek Hoiem,et al.  Learning CRFs Using Graph Cuts , 2008, ECCV.

[21]  H. Sebastian Seung,et al.  Maximin affinity learning of image segmentation , 2009, NIPS.

[22]  Vincent Ng,et al.  Graph-Cut-Based Anaphoricity Determination for Coreference Resolution , 2009, NAACL.

[23]  R. Castillo,et al.  A framework for evaluation of deformable image registration spatial accuracy using large landmark point sets , 2009, Physics in medicine and biology.

[24]  Thorsten Joachims,et al.  Cutting-plane training of structural SVMs , 2009, Machine Learning.

[25]  Ben Taskar,et al.  Structured Determinantal Point Processes , 2010, NIPS.

[26]  George Papandreou,et al.  Perturb-and-MAP random fields: Using discrete optimization to learn and sample from energy models , 2011, 2011 International Conference on Computer Vision.

[27]  Warren B. Powell,et al.  Dirichlet Process Mixtures of Generalized Linear Models , 2009, J. Mach. Learn. Res..

[28]  Justin Domke,et al.  Parameter learning with truncated message-passing , 2011, CVPR 2011.