Hello! Thank you for your contribution and code!
In this paper : One possible explanation is that the neural network approximation tends to exhibit similar frequencies as the leading eigenvectors of its NTK.
I want to know whether the NTK is computed at initialization?
Hello! Thank you for your contribution and code!
In this paper : One possible explanation is that the neural network approximation tends to exhibit similar frequencies as the leading eigenvectors of its NTK.
I want to know whether the NTK is computed at initialization?