Feldman's Memorization Lower Bound

A lower-bound result in machine learning theory concerning the cost of memorization, holding that the more frequently a sample occurs, the less additional generalization risk is required for a model to memorize it; new research shows this conclusion does not hold in general.