/zero-shot-unlearning

Official repo of the paper Zero-Shot Machine Unlearning accepted in IEEE Transactions on Information Forensics and Security

Primary LanguagePythonMIT LicenseMIT

Zero-Shot Machine Unlearning

Official repo of the paper Zero-Shot Machine Unlearning accepted in IEEE Transactions on Information Forensics and Security

Description

We introduce the problem statement of unlearning with no training samples and propose two solutions for the same. Unlearning quality has been evaluated through various metrics including membership inference attacks and inversion attacks. Also, a new metric called Anamnesis Index (AIN) has been introduced.

Paper

Zero-Shot Machine Unlearning

BibTex

@article{chundawat2023zero, title={Zero-shot machine unlearning}, author={Chundawat, Vikram S and Tarun, Ayush K and Mandal, Murari and Kankanhalli, Mohan}, journal={IEEE Transactions on Information Forensics and Security}, year={2023}, publisher={IEEE} }