A Generative Cramér-Rao Bound

The Cramér-Rao Bound (CRB) lets one compute the minimal mean-squared error that any unbiased estimator (algorithm) can achieve. Unfortunately, you need to know the probability density function (PDF) to compute the CRB. This work proposes a way to overcome this drawback. Maybe with this, the CRB gains more attention in the machine learning community - or even as much attention as it gets in the parameter estimation community.