Fine Tuning a Simulation-Driven Estimator