First of all, I’d like to say thanks and congratulations to Dr. Rackauckas and the team responsible for the DiffEqFlux module. This is an excellent module and I can see some really useful applications of this work. I have been working on getting to grips with it for a month or so, and despite being new to Julia, I have managed to get some simple applications working. Where I think there might be a nice use case is in the context of Dr. Rackauckas’s note on “Strategies to Avoid Local Minima” (https://diffeqflux.sciml.ai/dev/examples/local_minima/). I was hoping to write a program which would initially solve a differential equation system (e.g. Lotka-Volterra would do fine) with fixed initial conditions, u0=[a,b] say, and four parameters p0=[c,d,e,f]. The second part of the program would then start out with a small neural ODE network (minimally something like Chain(Dense(4,4,relu), Dense(4,4), p_new → solve(p_new, ODEProblem(lotka_volterra, u0, tspan, p_new) and try to learn the four parameters p_new, which should end up being close to [c,d,e,f], from the original ODE system, starting from a random guess of new parameters p_new=[g,h,i,j]. (Dr. Rackauckas has an example of doing something in the same spirit, but a quite different approach in the section “Training both the initial conditions and the parameters to start” in the aforementioned article.) I have tried several attempts of my own to do this, but luckily, it is unnecessary for me to include my clumsy code here, as there has been another attempt to do something very similar to this by a better programmer. This is shown in the question “predicting ODE parameters with DiffEqFlux” at stackoverflow (neural network - predicting ODE parameters with DiffEqFlux - Stack Overflow). Dr Rackauckas actually addressed this question, but it remains unsolved, according to the OP. I have also tried everything I can think of to make this code work, but again, unsuccessfully. I would hazard a guess that for someone very familiar with the current state of development of DiffEqFlux, it would not be a particularly onerous task to write a simple example. Could anyone help us out?