In this talk we consider the problem of learning dynamic functionals using a combination of static and dynamic random neural networks. We propose a novel class of functionals that can be viewed as a dynamic analogue of generalized Barron functions and that can be shown to contain a large class of practically relevant examples. In addition, such functionals can be approximated well by the considered random neural networks, as the derived error bounds show.

The talk is based on joint work with Lyudmila Grigoryeva and Juan-Pablo Ortega.