D. Soudry, E. Hoffer, M. Nacson, S. Gunasekar, and N. Srebro. (2017)cite arxiv:1710.10345Comment: Final JMLR version, with improved discussions over v3. Main improvements in journal version over conference version (v2 appeared in ICLR): We proved the measure zero case for main theorem (with implications for the rates), and the multi-class case.