natural parameters or covariance parameters for variational approximations

[thangbui]

Investigating ways to parameterise the variational approximations in the uncollapsed variational bound or approximate Power-EP energy, in particular q(u) = N(u; 0, LL^T) or q(u) = N_{natural}(u; \theta_1; Kuu^{-1} + \theta_2) -- which is faster or results in better learning curves etc?

See figure 1 here http://papers.nips.cc/paper/5559-decoupled-variational-gaussian-inference.pdf