LastiterateconvergenceofSGDforLeast-Squares intheInterpolationregime

Motivated by the recent successes of neural networks that have the ability to fit the data perfectlyand generalize well, we study the noiseless model in the fundamental least-squares setup.