😡 Mean Field Approximation

Mean field approximation is a type of variational inference that assumes the variational distribution to take on a specific form,

where there are no dependencies across hidden variables.

When we apply the 🧬 Evidence Lower Bound, we find that the optimal value for , denoted as , follows

where in the expectation denotes a marginalization over all for . We can thus optimize the ELBO via coordinate ascent, changing one dimension while keeping the others constant.