Model inversion
Model inversion: An attack that reconstructs features or examples from training data by exploiting model outputs. Memorization can facilitate reconstruction of images, text or sensitive attributes. Limiting exposed information, regularization and privacy evaluation reduce risk. Examples include Reconstructing a face, Extracting a memorized sequence, Inferring sensitive attributes.
In italiano: Model InversionLegge V · Overfitting, bias e collasso dei modelliMemorization can facilitate reconstruction of images, text or sensitive attributes. Limiting exposed information, regularization and privacy evaluation reduce risk.
Examples
- Reconstructing a face
- Extracting a memorized sequence
- Inferring sensitive attributes