how to fft2 with dlarray?

use in neural network training which need to define loss function

the loss function calaulated by the network input and output like this:

loss = mse(I,Y);

Y is the input of the network and I is the parameter which need fft2 through the output of the network

if use the extractdata(), the result can't be used to calculate gradient by

dlgradient(loss, net.Learnables)

in this situation the gradient only equal 0

Answer 1

[Matt J]

You haven't explained the difficulty you are experiencing, but I assume it is that an fft(), but not an fft2() command is available for dlarrays. However, you can build a 2D FFT out of 1D FFTs. Example:

X1=rand(400); %input to FFT
X2=dlarray(X1);

Y1=fft2(X1);

Y2=fft(fft(X2,[],2),[],1);

percentError = norm(Y1-extractdata(Y2),'inf')/norm(Y1,inf)*100

percentError = 1.3546e-13

[瑛杰]

Thank u for u reply!