To “ Multi - Way Blockmodels for Analyzing Coordinated High - Dimensional Responses ”

To carry out parameter estimation in the proposed multi-way blockmodel, we developed a variational EM algorithm in a similar fashion to (Airoldi, 2007; Jordan et al., 1999) with inference detailed as follows. N 1 and 1 ≤ k ≤ N 2. The latent variables X are assumed to follow distribution q(X). By Jensen's inequality, we obtain the following lower bound on the observed data log-likelihood: (A.1) log p(Y |Θ) ≥ E q [log p(Y, X|Θ)] − E q [log q(X)].