3 ms·
I recently published a new crazy convolution formula at https://arxiv.org/abs/1902.08982 https://arxiv.org/abs/1902.08982 (see last formulæ of the paper). I jus
by baruchel 8y ago
I recently published a new crazy convolution formula at https://arxiv.org/abs/1902.08982 https://arxiv.org/abs/1902.08982 (see last formulæ of the paper). I just submitted it to a journal and am waiting for the answer. Discussion is welcome!
- p1esk 8y agoCould you provide a PoC convolutional neural network code using your method? So we can then compare compute/memory intensity? p.s. I’m assuming of course that your method is functionally equivalent to the standard convolution. Is it?
- enriquto 8y agoFor easier reading and referencing you could specify the total big-O cost of your algorithm right in the beginning (or even in the abstract). And an indication on which range of input sizes it is likely to outperform karatsuba multiplication before being outperformed by the fft.