The field of machine unlearning, though still nascent, addresses exactly this problem. Could be useful in allowing ML models to satisfy data control regulations. Doing it well is still a significant challenge though.

→ View original post on X — @thegautamkamath,