Hacker News new | ask | show | jobs
by hikarudo 2090 days ago
How are CNNs directly based on FFTs? Sure you can use CNNs with FFT features, but in my experience this is not common.
1 comments

Convolutions are typically computed using FFTs.

https://en.wikipedia.org/wiki/Convolution_theorem

He is not wrong, convolutions between an image and a small kernel can be done faster by direct multiplication than by padding the kernel and performing FFT + iFFT. This is what tensor cores are aiming to do really fast. However, doing a convolution betwen an image and a kernel with the similar size is the general use case for the convolution theorem and is the thing that is currently implemented in VkFFT.