MUL is pretty well studied by now: given a trained model, can we "remove" all influence of some points in the training set? Naive way: retrain from scratch without those points. But this is slow. A number of faster methods have been proposed. 2/n

→ View original post on X — @thegautamkamath,