Generalizing the convolution -> matrix multiplication transformation for a single convolution filter, we can just as easily convert an entire conv layer with many filters into a matrix multiplication by just adding a new weight column for each filter. Then, each column of the result matrix will represent the convolution with the corresponding filter.
Generalizing the convolution -> matrix multiplication transformation for a single convolution filter, we can just as easily convert an entire conv layer with many filters into a matrix multiplication by just adding a new weight column for each filter. Then, each column of the result matrix will represent the convolution with the corresponding filter.