Understanding Deep Learning (Still) Requires Rethinking Generalization | March 2021 | Communications of the ACM

In this work, we presented a simple experimental framework for interrogating purported measures of generalization.