UCLA

This thesis develops a fast splitting method for estimating Gaussian Bayesian networks from observational data. In Bayesian network, the directed acyclic graph (DAG) structure is used to represent the causal interactions of the variables. Without using prior knowledge about the network structure, we learn the structure by cyclically augmenting new edges to the graph while forcing the structure to be a DAG. For the given structure we learn the optimal edge coefficients using Alternating Direction Method of Multipliers. Our approach is compared with the most recent method of CDDr, and the experiment result shows that our method outperforms CDDr significantly.