On: The Fourier transform is a neural network

[leecarraher]

Alt title, can a neural network learn the fast fourier transform?

[SuchAnonMuchWow → leecarraher]

the discrete Fourier transform, not the fast fourrier transform: the learned algorithm runs in O(n²) and not O(n*log n)

[leecarraher → SuchAnonMuchWow]

oh then this is just approximating the values of the linear transform. I figured at least they'd use mlp to learn each layer of the FFT butterfly network, and the corresponding weights. What would be of interest would be if you could extend this for n length fft using an RNN network