antkillerfarm.github.com
antkillerfarm.github.com copied to clipboard
antkillerfarm
Winograd Convolution
Hi, I came across your winograd convolution article at https://antkillerfarm.github.io/dl%20acceleration/2019/07/19/DL_acceleration.html
May I know what are ß0, ß1 and ß2 ?
Okay, I think I got it now.
For those interested, please see the mathematical process of lagrange interpolation
However, what I do not understand is why ß0 = 0, ß1 = 1 and ß2 = -1 ?
为何 当卷积核较大时，增加的加法数量以远超核大小的速度增长，最终会导致增加的加法所耗费的时间甚至超过节省下来的乘法所耗费的时间 ？
@promach x,x-1,x+1是线性无关的基，实际上任意x+i和x+j都是线性无关的，既然如此，为啥不用最简单的？ 至于你的第二个问题，我也是抄书。。。给不了证明。。。
@antkillerfarm
您对 More accurate Winograd/Cook/Toom F(4x4, 3x3) transforms , Error Analysis and Improving the Accuracy of Winograd Convolution for Deep Neural Networks 和 On improving the numerical stability of winograd convolutions 有什么高见吗 ？
至于你的第二个问题，我也是抄书。。。给不了证明。。。
是具体哪一本书呢 ？