Overview / Usage

'SGD Learns Over-Parameterized Networks that Provably Generalize on Linearly Separable Data' by Alon Brutzkus et al. This paper talks about generalization in deep learning. They prove convergence to a global minimum, and also prove a generalization bound that doesn't depend on the size of the network at all. Since most existing bounds depend on the size of the network, this is important due to the fact that we use noticeably overparameterized models that lead to vacuous generalization bounds.