[mlpack-git] [mlpack] implement CNN and LeNet1 (#405)

[Shangtong Zhang]

I don't think the paper is proper for this CNN. The value of fft(kernel) can be stored, but it seems I can't store fft(input). There may be different partition of input for different kernel as kernel size may vary. So for each convolution just one fft operation is reduced, there still exists 1 fft and 1 ifft. I think this optimization can't help fft-based convolution beat simple convolution upon small input and small kernel.

---
Reply to this email directly or view it on GitHub:
https://github.com/mlpack/mlpack/pull/405#issuecomment-82702654
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.cc.gatech.edu/pipermail/mlpack-git/attachments/20150317/25af5e80/attachment.html>