How can I get derivative between input and output parameters (dy/dx) in Artificial Neural Network

Hello, Dear all

I recently tring to run Physics informed Neural network in Matlab.

My objective is that, calculate derivatives between output and input and use this value into the physics. And Minimize the Loss function of it.

*Loss function = data based Loss function + Physics based Loss function = (target value-estimate value)^2+(physics)^2

for example, at the free fall movement ( input : time information // output : spatial information )

if we can get derivatives then we can get d^2x/dt^2 and use into the Governing law (F = ma) a = d^2x/dt^2 = g --> physics based Loss function = (d^2x/dt^2 - g)^2

x = [100 200 300 400 500]

t = [201 202 303 404 505]

net = feedforwardnet([3],'trainlm');

net = configure(net,x,t);

wb = getwb(net)

gwb = fpderiv('dperf_dwb',net,x,t)

But, I only can get derivatives for performance. not input and output relationship

so, How can i get derivatives?