CFE 2019: Start Registration
View Submission - CMStatistics
Title: Supermodular inequalities in hidden variable models Authors:  Anna Seigal - University of Oxford (United Kingdom) [presenting]
Guido Montufar - UCLA (United States)
Abstract: The implicit semi-algebraic description of a statistical model gives a membership test based on the signs of polynomials. We discuss supermodular inequalities, which take the form of signs of conditional independence statements. We focus on two graphical models with hidden variables, both on three binary observed variables. The semi-algebraic description of the models is given in terms of supermodular inequalities. We use this description to obtain a closed form expression for the maximum likelihood estimates, and discuss supermodular inequalities of larger models.