Model inversion attack

A model inversion attack is a privacy attack where the attacker is able to reconstruct the original samples that were used to train the synthetic model from the generated synthetic data set.