PROJECT TITLE :

Loop Calculus For Nonbinary Alphabets Using Concepts From Information Geometry

ABSTRACT:

The Bethe approximation could be a well-known approximation of the partition operate utilized in statistical physics. Recently, an equality relating the partition function and its Bethe approximation was obtained for graphical models with binary variables by Chertkov and Chernyak. During this equality, the multiplicative error within the Bethe approximation is represented as a weighted sum over all generalized loops within the graphical model. During this paper, the equality is generalized to graphical models with nonbinary alphabet using ideas from information geometry.