Sparse estimation of Linear Non-Gaussian Acyclic Model for Causal Discovery

Kazuharu Harada, Hironori Fujisawa

Abstract: Highlights•Existing algorithms for LiNGAM do not use sparseness and non-Gaussianity sufficiently.•Penalized likelihood-based approach enables us to estimate the model efficiently.•Solutions can be obtained by ADMM-based algorithm with some devices.•Our method exceeds the existing methods in numerical experiments and real data analysis.