Efficiently combining kernel methods and neural networks for operator learning leads to superior performance in nonlinear model reduction.
