Entire Relaxation Path for Maximum Entropy Problems

We discuss and analyze the problem of finding a distribution that minimizes the relative entropy to a prior distribution while satisfying max-norm constraints with respect to an observed distribution. This setting generalizes the classical maximum entropy problems as it relaxes the standard constraints on the observed values. We tackle the problem by introducing a re-parametrization in which the unknown distribution is distilled to a single scalar. We then describe a homotopy between the relaxation parameter and the distribution characterizing parameter. The homotopy also reveals an aesthetic symmetry between the prior distribution and the observed distribution. We then use the reformulated problem to describe a space and time efficient algorithm for tracking the entire relaxation path. Our derivations are based on a compact geometric view of the relaxation path as a piecewise linear function in a two dimensional space of the relaxation-characterization parameters. We demonstrate the usability of our approach by applying the problem to Zipfian distributions over a large alphabet.

[1]  Biing-Hwang Juang,et al.  Fundamentals of speech recognition , 1993, Prentice Hall signal processing series.

[2]  John Blitzer,et al.  Domain adaptation of natural language processing systems , 2008 .

[3]  Andrzej Stachurski,et al.  Parallel Optimization: Theory, Algorithms and Applications , 2000, Parallel Distributed Comput. Pract..

[4]  Frans M. J. Willems,et al.  The context-tree weighting method: basic properties , 1995, IEEE Trans. Inf. Theory.

[5]  Edwin T. Jaynes.  Prior Probabilities , 2010, Encyclopedia of Machine Learning.

[6]  Stephen P. Boyd,et al.  Convex Optimization , 2004, Algorithms and Theory of Computation Handbook.

[7]  R. Tibshirani,et al.  Least angle regression , 2004, math/0406456.

[8]  George Kingsley Zipf,et al.  Human behavior and the principle of least effort , 1949 .

[9]  John D. Lafferty,et al.  Inducing Features of Random Fields , 1995, IEEE Trans. Pattern Anal. Mach. Intell..

[10]  Miroslav Dudík,et al.  Maximum Entropy Density Estimation with Generalized Regularization and an Application to Species Distribution Modeling , 2007, J. Mach. Learn. Res..

[11]  Y. Censor,et al.  Parallel Optimization: Theory, Algorithms, and Applications , 1997 .

[12]  M. R. Osborne,et al.  On the LASSO and its Dual , 2000 .

[13]  Adam L. Berger,et al.  A Maximum Entropy Approach to Natural Language Processing , 1996, CL.