Bayesian learning in the absence of training data, by embedding model parameters into support of the likelihood: applications to astrophysics