Fig. 1
From: Module-based regularization improves Gaussian graphical models when observing noisy data
Comparing Standard and Modular GLASSO methods in detecting planted modular structure. The covariance matrix sampled from the Wishart distribution is noisy but with modular structure (a). With data sampled using this matrix, the GLASSO based on log-likelihood (Standard GLASSO) regularizes less than the GLASSO based on modular structure through Infomap’s codelength (Modular GLASSO) (b), which leads to the Standard GLASSO’s failure to identify any modular structure (c) while the Modular GLASSO successfully recovers the planted modular structure (d)