Title: On the attractive properties for estimating and generating distributions of special palindromic Ising models
Authors: Nanny Wermuth - Chalmers University of Technology (Sweden) [presenting]
Abstract: Palindromic Ising models are binary quadratic exponential distributions for marginally symmetric binary variables. Their joint probabilities have an additional, special symmetry since these probabilities listed in any lexicographic order remain unchanged after fully reversing its listed elements. The term palindromic stems from linguistics, where it was coined for sequences of characters which read forward and backward in the same way, like in the sentence `step on no pets'. Such binary distributions have been used first in statistical physics for special types of ferromagnetism, called there zero-mean Ising models. We concentrate on attractive properties for estimating and generating their distributions when all nodes in their undirected, simple and finite graphs can be ordered and interpreted to have at most two parent nodes.