Supplement material for " Walsh-Hadamard Variational Inference for Bayesian Deep Learning " Simone Rossi Sébastien Marmin

V). Therefore, we constrain the parameters such that Tr(V) = 1. The covariance matrices verify (see e.g. The Walsh-Hadamard matrix H is symmetric. The matrix A in Section 2.2 expresses the linear relationship between the weights W = S Equation (6) derives directly from block matrix and vector operations. As L is clearly lower triangular (even diagonal), let us proof that Q has orthogonal columns.