Anonymous Learning via Look-Alike Clustering: A Precise Analysis of Model Generalization