This talk has moved to Fall 2019 — Continuum approximations for wide neural networks and gradient descent.