Sparse Logistic Regression Learns All Discrete Pairwise Graphical Models

Neural Information Processing Systems 

We characterize the effectiveness of a classical algorithm for recovering the Markov graph of a general discrete pairwise graphical model from i.i.d.