EpochLex entry

Model Inversion

MOD-ul in-VER-zhun

A privacy attack that uses access to a trained model, its outputs, confidence information, gradients, or parameters to reconstruct representative or sensitive information about the data used to train the model.