A0947
Title: Approximations in the anisotropic Ising and 3-color Potts models for use in scene analysis
Authors: Alejandro Murua - University of Montreal (Canada) [presenting]
Ranjan Maitra - Iowa State University (United States)
Abstract: The Potts distribution is useful in many applications involving statistical modeling and inference. However, its normalization constant and its moments are intractable. This drawback has prevented a wider use of this model. The recent work is extended to fast numerical approximations to the homogeneous Ising model. Analytical approximations are supplied for the anisotropic Ising model and the 3-color Potts model. Simulation studies indicate good performance compared to Markov chain Monte Carlo methods and in a tiny fraction of the time. The methodology is illustrated with real applications and hypothesis testing.
