Apple Research: Not All Training Data Needs Expensive 'Unlearning' to Erase

Apple researchers find many training points barely affect a model, so skipping them could make privacy-driven data 'unlearning' much cheaper.