I am an attendee of the Frontiers of Deep Learning workshop at Simons Insitute for the Theory of Computing. There many well-known researchers here and just yesterday Yuanzhi Li gave a fantastic talk titled “Learning and Generalization in Over-parametrized Neural Networks, Going Beyond Kernels”. He presented a very simple explanation of what I believed to be very complicated and I encourage everyone interested in the topic to watch his presentation (should appear on youtube shortly).
The workshop website is here: https://simons.berkeley.edu/programs/dl2019
Update: the talk by Yuanzhi Li is available on Youtube: https://www.youtube.com/watch?v=NNPCk2gvTnI