The above equation corresponds to Kullback–Leibler divergence which is minimized when the estimated parameters are equal to the optimal parameters. The true
model θ u∗ is not known but could be estimated as the expectation over the posterior
distribution of the user’s model i.e. p(θ u|u).