HedgeCut: Maintaining Randomised Trees for Low-Latency Machine Unlearning

What is it about?

Personal control over data is a growing trend and this prominently includes the right to exclude your data from use by a company. When a user requests removal, the company has two choices with respect to machine learning-based systems. The first is to retrain all of the models that include that users data. The second is to modify production models by *un* learning the data associated with removed users. HedgeCut provides an efficient way to do that for an important class of models.

Featured Image