Return TransformedDistribution from DenseVariational Model

Question

I would like to quantify both epistemic and aleatoric uncertainty using dense variational layers to find the best parameters to a distribution layer.

Normally, this netwo