For DiffEqFlux, it might resemble adaptive control. Instead of finding cost at the final time. We could update parameters as if they are also states of the ODE along with normal states. Parameter derivatives can be from any of the optimizers. It might work or find locally acceptable parameters in each time step.