Approximate Machine Unlearning

In this chapter, we discuss approximate machine unlearning, a practical and widely adopted family of unlearning techniques that is particularly suitable for modern large-scale models where exact retraining is often computationally prohibitive. Unlike exact unlearning methods introduced in Chap. 2 , approximate unlearning aims to efficiently reduce or remove the influence of specified training samples from a trained model while preserving predictive utility of the retained data.