In this paper, it is said a new Fourier transform is up to 10x faster than previously known ones.
That's a serious claim, given that FFT can be already very fast
(for example, i know formulaes which require just a few additions & shifts per 8-pixel lines),
but i'm not able to judge the claim from the paper itself.
Still, it's MIT, so i expect it to be more than just BS.