Machine Unlearning

Machine Unlearning (MU) aims to enable machine learning models to forget or remove knowledge of certain data points in their training to meet needs such as privacy protection, legal requirements, or copyright protection.

There are two main strategies for machine forgetting: precise forgetting and approximate forgetting. Precise forgetting completely excludes the data that needs to be forgotten by retraining the model from scratch, but this method is computationally expensive. Approximate forgetting attempts to achieve forgetting by modifying the existing model, avoiding the high cost of retraining.