Reviews: Beyond normality: Learning sparse probabilistic graphical models in the non-Gaussian setting

This paper presents a method for identifying the independence structure of undirected probabilistic graphical models with continuous but non-Gaussian distributions, using SING, a novel iterative algorithm based on transport maps. The authors derive an estimate for the number of samples needed to recover the exact underlying graph structure with some probability and demonstrate empirically that SING can indeed recover this structure on two simple domains, where comparable methods that make invalid assumptions about the underlying data-generating process fail. The paper seems technically sound, with its claims and conclusions supported by existing and provided theoretical and empirical results. The authors mostly do a good job of justifying their approach, but do not discuss potential issues with the algorithm. For example, what is the complexity of SING?