High Dimensional Inference with Random Maximum A-Posteriori Perturbations

  title={High Dimensional Inference with Random Maximum A-Posteriori Perturbations},
  author={Tamir Hazan and Francesco Orabona and Anand D. Sarwate and Subhransu Maji and Tommi S. Jaakkola},
In this work we present a new approach for high-dimensional statistical inference that is based on optimization and random perturbations. This framework injects randomness to maximum a-posteriori (MAP) predictors by randomly perturbing its potential function. When the perturbations are of low dimension, sampling the perturb-max prediction is as efficient as MAP optimization. A classic result from extreme value statistics asserts that perturb-max operations generate unbiased samples from the… CONTINUE READING


Publications referenced by this paper.
Showing 1-10 of 61 references

The Concentration of Measure Phenomenon, ser. Mathematical Surveys and Monographs

M. Ledoux
American Mathematical Society, • 2001
View 7 Excerpts
Highly Influenced

Graphical models, exponential families, and variational inference

M. Wainwright, M. Jordan
Foundations and Trends in Machine Learning, vol. 1, no. 1-2, pp. 1–305, 2008. [Online]. Available: http://dx.doi.org/10.1561/2200000001 • 2008
View 4 Excerpts
Highly Influenced

On extensions of the Brunn-Minkowski and Prékopa-Leindler theorems, including February 12, 2016 DRAFT 44 inequalities for log concave functions, and with an application to the diffusion equation

H. J. Brascamp, E. H. Lieb
Journal of Functional Analysis, vol. 22, no. 4, pp. 366–389, August 1976. [Online]. Available: http://dx.doi.org/10.1016/0022-1236(76)90004-5 • 1976
View 5 Excerpts
Highly Influenced

Analysis and Geometry of Markov Diffusion Operators, ser

D. Bakry, I. Gentil, M. Ledoux
Grundlehren der mathematischen Wissenschaften. Switzerland: Springer International Publishing, • 2014
View 1 Excerpt

Dimensional variance inequalities of Brascamp–Lieb type and a local approach to dimensional Prékopas theorem

V. H. Nguyen
Journal of Functional Analysis, vol. 266, no. 2, pp. 931–955, January 2014. [Online]. Available: http://dx.doi.org/10.1016/j.jfa.2013.11.003 • 2014
View 2 Excerpts

Similar Papers

Loading similar papers…