Winograd's Convolution Theorem [Explained]

Convolution itself is a very hefty process, in most cases there is a lot of repetitive informations from one step of the window to the next.

This significant overlap gives us the intuition that there maybe more efficient methods, that could omit extra calculations and give us a more efficient convolution method.

This theorem which gives us this efficient method has a more mathematical descent and did not originate in Convolutional netwroks. We saw our first CNN is 1994, while we had this theorem...