Inference in Ising models (Q2405192)

The Ising spin glass is a discrete random field developed in statistical physics as a model for ferromagnetism, and is now used in statistics as a model for binary data with applications in spatial modeling, image processing, and neural networks. To describe the model, the authors suppose that the data is a vector of dependent \(\pm 1\) random variables and that a one-parameter exponential family with quadratic sufficient statistic models the dependence among the coordinates of this vector. It is shown that given a single realization from this model, the maximum pseudolikelihood estimate (MPLE) of the natural parameter is \(sqrt{a_n}\)-consistent at a point whenever the log-partition function has order \(a_n\) in a neighborhood of that point, as \(a_n\) tends to infinity. This gives consistency rates of the MPLE for ferromagnetic Ising models on general weighted graphs in all regimes. The authors also show that no consistent test, and hence no estimator, exists if the log-partition function remains bounded. Finally, the usefulness of the MPLE and the applicability of obtained results are illustrated.