Controllability of continuous networks and a kernel-based learning approximation

Michael Herty, Chiara Segala, Giuseppe Visconti

Residual deep neural networks are formulated as interacting particle systems leading to a description through neural differential equations, and, in the case of large input data, t…