Abstract: We introduce two approaches to graphical modeling for continuous and mixed data, using semi-parametric techniques that make weak assumptions compared with the default Gaussian graphical model. One approach is through semi-parametric extension of the Gaussian graphical model. Another approach is through semi-parametric extension of the exponential family graphical model. Both approaches can be viewed as adding structural regularization to a generic pairwise non-parametric Markov random field model. To fit these models, we will introduce a unified regularized rank-based estimation framework which naturally integrates statistical thinking and computational thinking. In terms of statistics, both methods achieve the optimal parametric rates of convergence. In terms of computation, both methods are as scalable as the best implemented parametric procedures. Such a "free lunch phenomenon" make them extremely attractive for large-scale applications. If time permits, we will also introduce some new research directions along this line of research.