We introduce new estimators for robust machine learning based on
median-of-means (MOM) estimators of the mean of real valued random variables.
These estimators achieve optimal rates of convergence under minimal assumptions
on the dataset. The dataset may also have been corrupted by out