A convergence result of a continuous model of deep learning via a Łojasiewicz--Simon inequality

Noboru Isobe

We study an idealized training process for deep neural networks in a continuous-depth, mean-field model in which each layer is parameterized by a probability measure on a Euclidean…