On the rate of convergence of an over-parametrized deep neural network regression estimate learned by gradient descent

Michael Kohler

Nonparametric regression with random design is considered. The $L_2$ error with integration with respect to the design measure is used as the error criterion. An over-parametrized…