Research question
How can a model forget specific training points efficiently and provably?
Key findings
- SISA training on sharded, sliced data bounds retraining cost for deletion requests - Exact unlearning is feasible for moderate deletion rates
Machine Unlearning
Introduced the SISA framework and the modern framing of machine unlearning for right-to-be-forgotten compliance.
Contribute
Spot an error, or know a record we should add? Members can propose additions and submit corrections.