/projects-amnesia

Code and experiments for our CIDR paper on "A Selection of Machine Learning Models that Can Forget User Data Very Fast"

Primary LanguageRustApache License 2.0Apache-2.0

Amnesia

To comply with laws, which protect the “right to be forgotten” and delete customer data from ML models (considered to be a lossy compressed version of customer data), Amnesia enables users to decrementally update trained ML models (for recommendation, classification, and regression), thereby forgetting the data of a customer without necessitating model retraining.

See our publication for details: https://ssc.io/pdf/amnesia-cidr.pdf