g****t
发帖数: 31659 | 1 say we had two time series
x(k)
y(k)
x(k)*y(-k) basically decided the transfer function between x and y.
For 2-d signals like imagine, it is the transfer function matrix or
solution of linear PDEs
What is the x*y*z*a*b*c*d... in high dimension? |
g*****y
发帖数: 7271 | 2 From frequency domain, it is simple:
it is XYZABCD...
It is harder when there are none linear operation after each convolution,
right?
【在 g****t 的大作中提到】
: say we had two time series
: x(k)
: y(k)
: x(k)*y(-k) basically decided the transfer function between x and y.
: For 2-d signals like imagine, it is the transfer function matrix or
: solution of linear PDEs
: What is the x*y*z*a*b*c*d... in high dimension?
|
g****t
发帖数: 31659 | 3 We cannot control the error.
【在 g*****y 的大作中提到】
: From frequency domain, it is simple:
: it is XYZABCD...
: It is harder when there are none linear operation after each convolution,
: right?
|
f*********g
发帖数: 25 | 4 如果convolution之后没有non-linear operation，根据Central-limit theorem，最终
的filter会变成Gaussian Filter。如果有non-linear operation，可以看一下这篇文章
https://www.di.ens.fr/~mallat/papiers/Bruna-Mallat-Pami-Scat.pdf
虽然它里面的non-linear operation用的是Fourier modulus针对的是complex value
filte和CNN的不一样,但是从single processing 和能量传播角度去看这个问题挺有意
思的 |
g****t
发帖数: 31659 | 5 Before finishing Einstein and Fourier's papers,
why we need to waste time on these papers?
I am old and I usually only checked modern papers' result and test data,
coding, 这些干货。
: 如果convolution之后没有non-linear operation，根据Central-limit
theorem
，最终
: 的filter会变成Gaussian Filter。如果有non-linear operation，可以
看一下
这篇文章
: https://www.di.ens.fr/~mallat/papiers/Bruna-Mallat-Pami-Scat.pdf
: 虽然它里面的non-linear operation用的是Fourier modulus针对的是
complex
value
: filte和CNN的不一样,但是从single processing 和能量传播角度去看这
个问题
挺有意
: 思的
【在 f*********g 的大作中提到】
: 如果convolution之后没有non-linear operation，根据Central-limit theorem，最终
: 的filter会变成Gaussian Filter。如果有non-linear operation，可以看一下这篇文章
: https://www.di.ens.fr/~mallat/papiers/Bruna-Mallat-Pami-Scat.pdf
: 虽然它里面的non-linear operation用的是Fourier modulus针对的是complex value
: filte和CNN的不一样,但是从single processing 和能量传播角度去看这个问题挺有意
: 思的
|
